xref: /illumos-gate/usr/src/uts/i86pc/os/mlsetup.c (revision 80b758da23abee3ef0e550f533aada9ce3b1b01f)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2012 Gary Mills
23  *
24  * Copyright (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2011 by Delphix. All rights reserved.
26  * Copyright 2019 Joyent, Inc.
27  * Copyright 2024 Oxide Computer Company
28  */
29 /*
30  * Copyright (c) 2010, Intel Corporation.
31  * All rights reserved.
32  */
33 
34 #include <sys/types.h>
35 #include <sys/sysmacros.h>
36 #include <sys/disp.h>
37 #include <sys/promif.h>
38 #include <sys/clock.h>
39 #include <sys/cpuvar.h>
40 #include <sys/stack.h>
41 #include <vm/as.h>
42 #include <vm/hat.h>
43 #include <sys/reboot.h>
44 #include <sys/avintr.h>
45 #include <sys/vtrace.h>
46 #include <sys/proc.h>
47 #include <sys/thread.h>
48 #include <sys/cpupart.h>
49 #include <sys/pset.h>
50 #include <sys/copyops.h>
51 #include <sys/pg.h>
52 #include <sys/disp.h>
53 #include <sys/debug.h>
54 #include <sys/sunddi.h>
55 #include <sys/x86_archext.h>
56 #include <sys/privregs.h>
57 #include <sys/machsystm.h>
58 #include <sys/ontrap.h>
59 #include <sys/bootconf.h>
60 #include <sys/boot_console.h>
61 #include <sys/kdi_machimpl.h>
62 #include <sys/archsystm.h>
63 #include <sys/promif.h>
64 #include <sys/pci_cfgspace.h>
65 #include <sys/apic.h>
66 #include <sys/apic_common.h>
67 #include <sys/bootvfs.h>
68 #include <sys/tsc.h>
69 #include <sys/smt.h>
70 #ifdef __xpv
71 #include <sys/hypervisor.h>
72 #else
73 #include <sys/xpv_support.h>
74 #endif
75 
76 /*
77  * some globals for patching the result of cpuid
78  * to solve problems w/ creative cpu vendors
79  */
80 
81 extern uint32_t cpuid_feature_ecx_include;
82 extern uint32_t cpuid_feature_ecx_exclude;
83 extern uint32_t cpuid_feature_edx_include;
84 extern uint32_t cpuid_feature_edx_exclude;
85 
86 nmi_action_t nmi_action = NMI_ACTION_UNSET;
87 
88 /*
89  * Set console mode
90  */
91 static void
92 set_console_mode(uint8_t val)
93 {
94 	struct bop_regs rp = {0};
95 
96 	rp.eax.byte.ah = 0x0;
97 	rp.eax.byte.al = val;
98 	rp.ebx.word.bx = 0x0;
99 
100 	BOP_DOINT(bootops, 0x10, &rp);
101 }
102 
103 
104 /*
105  * Setup routine called right before main(). Interposing this function
106  * before main() allows us to call it in a machine-independent fashion.
107  */
108 void
109 mlsetup(struct regs *rp)
110 {
111 	u_longlong_t prop_value;
112 	char prop_str[BP_MAX_STRLEN];
113 	extern struct classfuncs sys_classfuncs;
114 	extern disp_t cpu0_disp;
115 	extern char t0stack[];
116 	extern int post_fastreboot;
117 	extern uint64_t plat_dr_options;
118 
119 	ASSERT_STACK_ALIGNED();
120 
121 	/*
122 	 * initialize cpu_self
123 	 */
124 	cpu[0]->cpu_self = cpu[0];
125 
126 #if defined(__xpv)
127 	/*
128 	 * Point at the hypervisor's virtual cpu structure
129 	 */
130 	cpu[0]->cpu_m.mcpu_vcpu_info = &HYPERVISOR_shared_info->vcpu_info[0];
131 #endif
132 
133 	/*
134 	 * check if we've got special bits to clear or set
135 	 * when checking cpu features
136 	 */
137 
138 	if (bootprop_getval("cpuid_feature_ecx_include", &prop_value) != 0)
139 		cpuid_feature_ecx_include = 0;
140 	else
141 		cpuid_feature_ecx_include = (uint32_t)prop_value;
142 
143 	if (bootprop_getval("cpuid_feature_ecx_exclude", &prop_value) != 0)
144 		cpuid_feature_ecx_exclude = 0;
145 	else
146 		cpuid_feature_ecx_exclude = (uint32_t)prop_value;
147 
148 	if (bootprop_getval("cpuid_feature_edx_include", &prop_value) != 0)
149 		cpuid_feature_edx_include = 0;
150 	else
151 		cpuid_feature_edx_include = (uint32_t)prop_value;
152 
153 	if (bootprop_getval("cpuid_feature_edx_exclude", &prop_value) != 0)
154 		cpuid_feature_edx_exclude = 0;
155 	else
156 		cpuid_feature_edx_exclude = (uint32_t)prop_value;
157 
158 #if !defined(__xpv)
159 	if (bootprop_getstr("nmi", prop_str, sizeof (prop_str)) == 0) {
160 		if (strcmp(prop_str, "ignore") == 0) {
161 			nmi_action = NMI_ACTION_IGNORE;
162 		} else if (strcmp(prop_str, "panic") == 0) {
163 			nmi_action = NMI_ACTION_PANIC;
164 		} else if (strcmp(prop_str, "kmdb") == 0) {
165 			nmi_action = NMI_ACTION_KMDB;
166 		} else {
167 			prom_printf("unix: ignoring unknown nmi=%s\n",
168 			    prop_str);
169 		}
170 	}
171 
172 	/*
173 	 * Check to see if KPTI has been explicitly enabled or disabled.
174 	 * We have to check this before init_desctbls().
175 	 */
176 	if (bootprop_getval("kpti", &prop_value) == 0) {
177 		kpti_enable = (uint64_t)(prop_value == 1);
178 		prom_printf("unix: forcing kpti to %s due to boot argument\n",
179 		    (kpti_enable == 1) ? "ON" : "OFF");
180 	} else {
181 		kpti_enable = 1;
182 	}
183 
184 	if (bootprop_getval("pcid", &prop_value) == 0 && prop_value == 0) {
185 		prom_printf("unix: forcing pcid to OFF due to boot argument\n");
186 		x86_use_pcid = 0;
187 	} else if (kpti_enable != 1) {
188 		x86_use_pcid = 0;
189 	}
190 
191 	/*
192 	 * While we don't need to check this until later, we might as well do it
193 	 * here.
194 	 */
195 	if (bootprop_getstr("smt_enabled", prop_str, sizeof (prop_str)) == 0) {
196 		if (strcasecmp(prop_str, "false") == 0 ||
197 		    strcmp(prop_str, "0") == 0)
198 			smt_boot_disable = 1;
199 	}
200 
201 #endif
202 
203 	/*
204 	 * Initialize idt0, gdt0, ldt0_default, ktss0 and dftss.
205 	 */
206 	init_desctbls();
207 
208 	/*
209 	 * initialize t0
210 	 */
211 	t0.t_stk = (caddr_t)rp - MINFRAME;
212 	t0.t_stkbase = t0stack;
213 	t0.t_pri = maxclsyspri - 3;
214 	t0.t_schedflag = TS_LOAD | TS_DONT_SWAP;
215 	t0.t_procp = &p0;
216 	t0.t_plockp = &p0lock.pl_lock;
217 	t0.t_lwp = &lwp0;
218 	t0.t_forw = &t0;
219 	t0.t_back = &t0;
220 	t0.t_next = &t0;
221 	t0.t_prev = &t0;
222 	t0.t_cpu = cpu[0];
223 	t0.t_disp_queue = &cpu0_disp;
224 	t0.t_bind_cpu = PBIND_NONE;
225 	t0.t_bind_pset = PS_NONE;
226 	t0.t_bindflag = (uchar_t)default_binding_mode;
227 	t0.t_cpupart = &cp_default;
228 	t0.t_clfuncs = &sys_classfuncs.thread;
229 	t0.t_copyops = NULL;
230 	THREAD_ONPROC(&t0, CPU);
231 
232 	lwp0.lwp_thread = &t0;
233 	lwp0.lwp_regs = (void *)rp;
234 	lwp0.lwp_procp = &p0;
235 	t0.t_tid = p0.p_lwpcnt = p0.p_lwprcnt = p0.p_lwpid = 1;
236 
237 	p0.p_exec = NULL;
238 	p0.p_stat = SRUN;
239 	p0.p_flag = SSYS;
240 	p0.p_tlist = &t0;
241 	p0.p_stksize = 2*PAGESIZE;
242 	p0.p_stkpageszc = 0;
243 	p0.p_as = &kas;
244 	p0.p_lockp = &p0lock;
245 	p0.p_brkpageszc = 0;
246 	p0.p_t1_lgrpid = LGRP_NONE;
247 	p0.p_tr_lgrpid = LGRP_NONE;
248 	psecflags_default(&p0.p_secflags);
249 
250 	sigorset(&p0.p_ignore, &ignoredefault);
251 
252 	CPU->cpu_thread = &t0;
253 	bzero(&cpu0_disp, sizeof (disp_t));
254 	CPU->cpu_disp = &cpu0_disp;
255 	CPU->cpu_disp->disp_cpu = CPU;
256 	CPU->cpu_dispthread = &t0;
257 	CPU->cpu_idle_thread = &t0;
258 	CPU->cpu_flags = CPU_READY | CPU_RUNNING | CPU_EXISTS | CPU_ENABLE;
259 	CPU->cpu_dispatch_pri = t0.t_pri;
260 
261 	CPU->cpu_id = 0;
262 
263 	CPU->cpu_pri = 12;		/* initial PIL for the boot CPU */
264 
265 	/*
266 	 * Ensure that we have set the necessary feature bits before setting up
267 	 * PCI config space access.
268 	 */
269 	cpuid_execpass(cpu[0], CPUID_PASS_PRELUDE, x86_featureset);
270 
271 	/*
272 	 * lgrp_init() and possibly cpuid_pass1() need PCI config
273 	 * space access
274 	 */
275 #if defined(__xpv)
276 	if (DOMAIN_IS_INITDOMAIN(xen_info))
277 		pci_cfgspace_init();
278 #else
279 	pci_cfgspace_init();
280 	/*
281 	 * Initialize the platform type from CPU 0 to ensure that
282 	 * determine_platform() is only ever called once.
283 	 */
284 	determine_platform();
285 #endif
286 
287 	/*
288 	 * While the BIOS may have already applied some microcode updates, we
289 	 * may have more recent updates available that we'd like to apply.  The
290 	 * application of said microcode may end up resulting in architecturally
291 	 * visible changes (e.g., changed MSR or CPUID bits) which we'd like to
292 	 * have in place before we start querying the CPU for its capabilities.
293 	 * So we first run the IDENT pass to determine the specific CPU vendor,
294 	 * model, rev etc., fill out cpu_ucode_info and update the microcode, if
295 	 * necessary.
296 	 */
297 	cpuid_execpass(cpu[0], CPUID_PASS_IDENT, NULL);
298 	ucode_init();
299 	ucode_check_boot();
300 
301 	/*
302 	 * Now we're ready to run the BASIC cpuid pass.
303 	 *
304 	 * The x86_featureset is initialized here based on the capabilities of
305 	 * the boot CPU.  Note that if we choose to support CPUs that have
306 	 * different feature sets (at which point we would almost certainly want
307 	 * to set the feature bits to correspond to the feature minimum) this
308 	 * value may be altered.
309 	 */
310 	cpuid_execpass(cpu[0], CPUID_PASS_BASIC, x86_featureset);
311 
312 #if !defined(__xpv)
313 	if ((get_hwenv() & HW_XEN_HVM) != 0)
314 		xen_hvm_init();
315 
316 	/*
317 	 * Before we do anything with the TSCs, we need to work around
318 	 * Intel erratum BT81.  On some CPUs, warm reset does not
319 	 * clear the TSC.  If we are on such a CPU, we will clear TSC ourselves
320 	 * here.  Other CPUs will clear it when we boot them later, and the
321 	 * resulting skew will be handled by tsc_sync_master()/_slave();
322 	 * note that such skew already exists and has to be handled anyway.
323 	 *
324 	 * We do this only on metal.  This same problem can occur with a
325 	 * hypervisor that does not happen to virtualise a TSC that starts from
326 	 * zero, regardless of CPU type; however, we do not expect hypervisors
327 	 * that do not virtualise TSC that way to handle writes to TSC
328 	 * correctly, either.
329 	 */
330 	if (get_hwenv() == HW_NATIVE &&
331 	    cpuid_getvendor(CPU) == X86_VENDOR_Intel &&
332 	    cpuid_getfamily(CPU) == 6 &&
333 	    (cpuid_getmodel(CPU) == 0x2d || cpuid_getmodel(CPU) == 0x3e) &&
334 	    is_x86_feature(x86_featureset, X86FSET_TSC)) {
335 		(void) wrmsr(REG_TSC, 0UL);
336 	}
337 
338 	/*
339 	 * Patch the tsc_read routine with appropriate set of instructions,
340 	 * depending on the processor family and architecure, to read the
341 	 * time-stamp counter while ensuring no out-of-order execution.
342 	 * Patch it while the kernel text is still writable.
343 	 *
344 	 * The Xen hypervisor does not correctly report whether rdtscp is
345 	 * supported or not, so we must assume that it is not.
346 	 */
347 	if ((get_hwenv() & HW_XEN_HVM) == 0 &&
348 	    is_x86_feature(x86_featureset, X86FSET_TSCP)) {
349 		patch_tsc_read(TSC_TSCP);
350 	} else if (is_x86_feature(x86_featureset, X86FSET_LFENCE_SER)) {
351 		ASSERT(is_x86_feature(x86_featureset, X86FSET_SSE2));
352 		patch_tsc_read(TSC_RDTSC_LFENCE);
353 	}
354 
355 #endif	/* !__xpv */
356 
357 
358 #if !defined(__xpv)
359 	patch_memops(cpuid_getvendor(CPU));
360 #endif	/* !__xpv */
361 
362 #if !defined(__xpv)
363 	/* XXPV	what, if anything, should be dorked with here under xen? */
364 
365 	/*
366 	 * While we're thinking about the TSC, let's set up %cr4 so that
367 	 * userland can issue rdtsc, and initialize the TSC_AUX value
368 	 * (the cpuid) for the rdtscp instruction on appropriately
369 	 * capable hardware.
370 	 */
371 	if (is_x86_feature(x86_featureset, X86FSET_TSC))
372 		setcr4(getcr4() & ~CR4_TSD);
373 
374 	if (is_x86_feature(x86_featureset, X86FSET_TSCP))
375 		(void) wrmsr(MSR_AMD_TSCAUX, 0);
376 
377 	/*
378 	 * Let's get the other %cr4 stuff while we're here. Note, we defer
379 	 * enabling CR4_SMAP until startup_end(); however, that's importantly
380 	 * before we start other CPUs. That ensures that it will be synced out
381 	 * to other CPUs.
382 	 */
383 	if (is_x86_feature(x86_featureset, X86FSET_DE))
384 		setcr4(getcr4() | CR4_DE);
385 
386 	if (is_x86_feature(x86_featureset, X86FSET_SMEP))
387 		setcr4(getcr4() | CR4_SMEP);
388 #endif /* __xpv */
389 
390 
391 	/*
392 	 * Initialize thread/cpu microstate accounting
393 	 */
394 	init_mstate(&t0, LMS_SYSTEM);
395 	init_cpu_mstate(CPU, CMS_SYSTEM);
396 
397 	/*
398 	 * Initialize lists of available and active CPUs.
399 	 */
400 	cpu_list_init(CPU);
401 
402 	pg_cpu_bootstrap(CPU);
403 
404 	/*
405 	 * Now that we have taken over the GDT, IDT and have initialized
406 	 * active CPU list it's time to inform kmdb if present.
407 	 */
408 	if (boothowto & RB_DEBUG)
409 		kdi_idt_sync();
410 
411 	if (BOP_GETPROPLEN(bootops, "efi-systab") < 0) {
412 		/*
413 		 * In BIOS system, explicitly set console to text mode (0x3)
414 		 * if this is a boot post Fast Reboot, and the console is set
415 		 * to CONS_SCREEN_TEXT.
416 		 */
417 		if (post_fastreboot &&
418 		    boot_console_type(NULL) == CONS_SCREEN_TEXT) {
419 			set_console_mode(0x3);
420 		}
421 	}
422 
423 	/*
424 	 * If requested (boot -d) drop into kmdb.
425 	 *
426 	 * This must be done after cpu_list_init() on the 64-bit kernel
427 	 * since taking a trap requires that we re-compute gsbase based
428 	 * on the cpu list.
429 	 */
430 	if (boothowto & RB_DEBUGENTER)
431 		kmdb_enter();
432 
433 	cpu_vm_data_init(CPU);
434 
435 	rp->r_fp = 0;	/* terminate kernel stack traces! */
436 
437 	prom_init("kernel", (void *)NULL);
438 
439 	/* User-set option overrides firmware value. */
440 	if (bootprop_getval(PLAT_DR_OPTIONS_NAME, &prop_value) == 0) {
441 		plat_dr_options = (uint64_t)prop_value;
442 	}
443 #if defined(__xpv)
444 	/* No support of DR operations on xpv */
445 	plat_dr_options = 0;
446 #else	/* __xpv */
447 	/* Flag PLAT_DR_FEATURE_ENABLED should only be set by DR driver. */
448 	plat_dr_options &= ~PLAT_DR_FEATURE_ENABLED;
449 #endif	/* __xpv */
450 
451 	/*
452 	 * Get value of "plat_dr_physmax" boot option.
453 	 * It overrides values calculated from MSCT or SRAT table.
454 	 */
455 	if (bootprop_getval(PLAT_DR_PHYSMAX_NAME, &prop_value) == 0) {
456 		plat_dr_physmax = ((uint64_t)prop_value) >> PAGESHIFT;
457 	}
458 
459 	/* Get value of boot_ncpus. */
460 	if (bootprop_getval(BOOT_NCPUS_NAME, &prop_value) != 0) {
461 		boot_ncpus = NCPU;
462 	} else {
463 		boot_ncpus = (int)prop_value;
464 		if (boot_ncpus <= 0 || boot_ncpus > NCPU)
465 			boot_ncpus = NCPU;
466 	}
467 
468 	/*
469 	 * Set max_ncpus and boot_max_ncpus to boot_ncpus if platform doesn't
470 	 * support CPU DR operations.
471 	 */
472 	if (plat_dr_support_cpu() == 0) {
473 		max_ncpus = boot_max_ncpus = boot_ncpus;
474 	} else {
475 		if (bootprop_getval(PLAT_MAX_NCPUS_NAME, &prop_value) != 0) {
476 			max_ncpus = NCPU;
477 		} else {
478 			max_ncpus = (int)prop_value;
479 			if (max_ncpus <= 0 || max_ncpus > NCPU) {
480 				max_ncpus = NCPU;
481 			}
482 			if (boot_ncpus > max_ncpus) {
483 				boot_ncpus = max_ncpus;
484 			}
485 		}
486 
487 		if (bootprop_getval(BOOT_MAX_NCPUS_NAME, &prop_value) != 0) {
488 			boot_max_ncpus = boot_ncpus;
489 		} else {
490 			boot_max_ncpus = (int)prop_value;
491 			if (boot_max_ncpus <= 0 || boot_max_ncpus > NCPU) {
492 				boot_max_ncpus = boot_ncpus;
493 			} else if (boot_max_ncpus > max_ncpus) {
494 				boot_max_ncpus = max_ncpus;
495 			}
496 		}
497 	}
498 
499 	/*
500 	 * Initialize the lgrp framework
501 	 */
502 	lgrp_init(LGRP_INIT_STAGE1);
503 
504 	if (boothowto & RB_HALT) {
505 		prom_printf("unix: kernel halted by -h flag\n");
506 		prom_enter_mon();
507 	}
508 
509 	ASSERT_STACK_ALIGNED();
510 
511 	if (workaround_errata(CPU) != 0)
512 		panic("critical workaround(s) missing for boot cpu");
513 }
514 
515 
516 void
517 mach_modpath(char *path, const char *filename)
518 {
519 	/*
520 	 * Construct the directory path from the filename.
521 	 */
522 
523 	int len;
524 	char *p;
525 	const char isastr[] = "/amd64";
526 	size_t isalen = strlen(isastr);
527 
528 	len = strlen(SYSTEM_BOOT_PATH "/kernel");
529 	(void) strcpy(path, SYSTEM_BOOT_PATH "/kernel ");
530 	path += len + 1;
531 
532 	if ((p = strrchr(filename, '/')) == NULL)
533 		return;
534 
535 	while (p > filename && *(p - 1) == '/')
536 		p--;	/* remove trailing '/' characters */
537 	if (p == filename)
538 		p++;	/* so "/" -is- the modpath in this case */
539 
540 	/*
541 	 * Remove optional isa-dependent directory name - the module
542 	 * subsystem will put this back again (!)
543 	 */
544 	len = p - filename;
545 	if (len > isalen &&
546 	    strncmp(&filename[len - isalen], isastr, isalen) == 0)
547 		p -= isalen;
548 
549 	/*
550 	 * "/platform/mumblefrotz" + " " + MOD_DEFPATH
551 	 */
552 	len += (p - filename) + 1 + strlen(MOD_DEFPATH) + 1;
553 	(void) strncpy(path, filename, p - filename);
554 }
555