xref: /illumos-gate/usr/src/uts/i86pc/os/mlsetup.c (revision e59325b74ca0760391c4b4d3acd8d3294ffc521a)
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  */
28 /*
29  * Copyright (c) 2010, Intel Corporation.
30  * All rights reserved.
31  */
32 
33 #include <sys/types.h>
34 #include <sys/sysmacros.h>
35 #include <sys/disp.h>
36 #include <sys/promif.h>
37 #include <sys/clock.h>
38 #include <sys/cpuvar.h>
39 #include <sys/stack.h>
40 #include <vm/as.h>
41 #include <vm/hat.h>
42 #include <sys/reboot.h>
43 #include <sys/avintr.h>
44 #include <sys/vtrace.h>
45 #include <sys/proc.h>
46 #include <sys/thread.h>
47 #include <sys/cpupart.h>
48 #include <sys/pset.h>
49 #include <sys/copyops.h>
50 #include <sys/pg.h>
51 #include <sys/disp.h>
52 #include <sys/debug.h>
53 #include <sys/sunddi.h>
54 #include <sys/x86_archext.h>
55 #include <sys/privregs.h>
56 #include <sys/machsystm.h>
57 #include <sys/ontrap.h>
58 #include <sys/bootconf.h>
59 #include <sys/boot_console.h>
60 #include <sys/kdi_machimpl.h>
61 #include <sys/archsystm.h>
62 #include <sys/promif.h>
63 #include <sys/pci_cfgspace.h>
64 #include <sys/apic.h>
65 #include <sys/apic_common.h>
66 #include <sys/bootvfs.h>
67 #include <sys/tsc.h>
68 #ifdef __xpv
69 #include <sys/hypervisor.h>
70 #else
71 #include <sys/xpv_support.h>
72 #endif
73 
74 /*
75  * some globals for patching the result of cpuid
76  * to solve problems w/ creative cpu vendors
77  */
78 
79 extern uint32_t cpuid_feature_ecx_include;
80 extern uint32_t cpuid_feature_ecx_exclude;
81 extern uint32_t cpuid_feature_edx_include;
82 extern uint32_t cpuid_feature_edx_exclude;
83 
84 nmi_action_t nmi_action = NMI_ACTION_UNSET;
85 
86 /*
87  * Set console mode
88  */
89 static void
90 set_console_mode(uint8_t val)
91 {
92 	struct bop_regs rp = {0};
93 
94 	rp.eax.byte.ah = 0x0;
95 	rp.eax.byte.al = val;
96 	rp.ebx.word.bx = 0x0;
97 
98 	BOP_DOINT(bootops, 0x10, &rp);
99 }
100 
101 
102 /*
103  * Setup routine called right before main(). Interposing this function
104  * before main() allows us to call it in a machine-independent fashion.
105  */
106 void
107 mlsetup(struct regs *rp)
108 {
109 	u_longlong_t prop_value;
110 	char prop_str[BP_MAX_STRLEN];
111 	extern struct classfuncs sys_classfuncs;
112 	extern disp_t cpu0_disp;
113 	extern char t0stack[];
114 	extern int post_fastreboot;
115 	extern uint64_t plat_dr_options;
116 
117 	ASSERT_STACK_ALIGNED();
118 
119 	/*
120 	 * initialize cpu_self
121 	 */
122 	cpu[0]->cpu_self = cpu[0];
123 
124 #if defined(__xpv)
125 	/*
126 	 * Point at the hypervisor's virtual cpu structure
127 	 */
128 	cpu[0]->cpu_m.mcpu_vcpu_info = &HYPERVISOR_shared_info->vcpu_info[0];
129 #endif
130 
131 	/*
132 	 * check if we've got special bits to clear or set
133 	 * when checking cpu features
134 	 */
135 
136 	if (bootprop_getval("cpuid_feature_ecx_include", &prop_value) != 0)
137 		cpuid_feature_ecx_include = 0;
138 	else
139 		cpuid_feature_ecx_include = (uint32_t)prop_value;
140 
141 	if (bootprop_getval("cpuid_feature_ecx_exclude", &prop_value) != 0)
142 		cpuid_feature_ecx_exclude = 0;
143 	else
144 		cpuid_feature_ecx_exclude = (uint32_t)prop_value;
145 
146 	if (bootprop_getval("cpuid_feature_edx_include", &prop_value) != 0)
147 		cpuid_feature_edx_include = 0;
148 	else
149 		cpuid_feature_edx_include = (uint32_t)prop_value;
150 
151 	if (bootprop_getval("cpuid_feature_edx_exclude", &prop_value) != 0)
152 		cpuid_feature_edx_exclude = 0;
153 	else
154 		cpuid_feature_edx_exclude = (uint32_t)prop_value;
155 
156 #if !defined(__xpv)
157 	if (bootprop_getstr("nmi", prop_str, sizeof (prop_str)) == 0) {
158 		if (strcmp(prop_str, "ignore") == 0) {
159 			nmi_action = NMI_ACTION_IGNORE;
160 		} else if (strcmp(prop_str, "panic") == 0) {
161 			nmi_action = NMI_ACTION_PANIC;
162 		} else if (strcmp(prop_str, "kmdb") == 0) {
163 			nmi_action = NMI_ACTION_KMDB;
164 		} else {
165 			prom_printf("unix: ignoring unknown nmi=%s\n",
166 			    prop_str);
167 		}
168 	}
169 
170 	/*
171 	 * Check to see if KPTI has been explicitly enabled or disabled.
172 	 * We have to check this before init_desctbls().
173 	 */
174 	if (bootprop_getval("kpti", &prop_value) == 0) {
175 		kpti_enable = (uint64_t)(prop_value == 1);
176 		prom_printf("unix: forcing kpti to %s due to boot argument\n",
177 		    (kpti_enable == 1) ? "ON" : "OFF");
178 	} else {
179 		kpti_enable = 1;
180 	}
181 
182 	if (bootprop_getval("pcid", &prop_value) == 0 && prop_value == 0) {
183 		prom_printf("unix: forcing pcid to OFF due to boot argument\n");
184 		x86_use_pcid = 0;
185 	} else if (kpti_enable != 1) {
186 		x86_use_pcid = 0;
187 	}
188 #endif
189 
190 	/*
191 	 * Initialize idt0, gdt0, ldt0_default, ktss0 and dftss.
192 	 */
193 	init_desctbls();
194 
195 	/*
196 	 * lgrp_init() and possibly cpuid_pass1() need PCI config
197 	 * space access
198 	 */
199 #if defined(__xpv)
200 	if (DOMAIN_IS_INITDOMAIN(xen_info))
201 		pci_cfgspace_init();
202 #else
203 	pci_cfgspace_init();
204 	/*
205 	 * Initialize the platform type from CPU 0 to ensure that
206 	 * determine_platform() is only ever called once.
207 	 */
208 	determine_platform();
209 #endif
210 
211 	/*
212 	 * The first lightweight pass (pass0) through the cpuid data
213 	 * was done in locore before mlsetup was called.  Do the next
214 	 * pass in C code.
215 	 *
216 	 * The x86_featureset is initialized here based on the capabilities
217 	 * of the boot CPU.  Note that if we choose to support CPUs that have
218 	 * different feature sets (at which point we would almost certainly
219 	 * want to set the feature bits to correspond to the feature
220 	 * minimum) this value may be altered.
221 	 */
222 	cpuid_pass1(cpu[0], x86_featureset);
223 
224 #if !defined(__xpv)
225 	if ((get_hwenv() & HW_XEN_HVM) != 0)
226 		xen_hvm_init();
227 
228 	/*
229 	 * Before we do anything with the TSCs, we need to work around
230 	 * Intel erratum BT81.  On some CPUs, warm reset does not
231 	 * clear the TSC.  If we are on such a CPU, we will clear TSC ourselves
232 	 * here.  Other CPUs will clear it when we boot them later, and the
233 	 * resulting skew will be handled by tsc_sync_master()/_slave();
234 	 * note that such skew already exists and has to be handled anyway.
235 	 *
236 	 * We do this only on metal.  This same problem can occur with a
237 	 * hypervisor that does not happen to virtualise a TSC that starts from
238 	 * zero, regardless of CPU type; however, we do not expect hypervisors
239 	 * that do not virtualise TSC that way to handle writes to TSC
240 	 * correctly, either.
241 	 */
242 	if (get_hwenv() == HW_NATIVE &&
243 	    cpuid_getvendor(CPU) == X86_VENDOR_Intel &&
244 	    cpuid_getfamily(CPU) == 6 &&
245 	    (cpuid_getmodel(CPU) == 0x2d || cpuid_getmodel(CPU) == 0x3e) &&
246 	    is_x86_feature(x86_featureset, X86FSET_TSC)) {
247 		(void) wrmsr(REG_TSC, 0UL);
248 	}
249 
250 	/*
251 	 * Patch the tsc_read routine with appropriate set of instructions,
252 	 * depending on the processor family and architecure, to read the
253 	 * time-stamp counter while ensuring no out-of-order execution.
254 	 * Patch it while the kernel text is still writable.
255 	 *
256 	 * Note: tsc_read is not patched for intel processors whose family
257 	 * is >6 and for amd whose family >f (in case they don't support rdtscp
258 	 * instruction, unlikely). By default tsc_read will use cpuid for
259 	 * serialization in such cases. The following code needs to be
260 	 * revisited if intel processors of family >= f retains the
261 	 * instruction serialization nature of mfence instruction.
262 	 * Note: tsc_read is not patched for x86 processors which do
263 	 * not support "mfence". By default tsc_read will use cpuid for
264 	 * serialization in such cases.
265 	 *
266 	 * The Xen hypervisor does not correctly report whether rdtscp is
267 	 * supported or not, so we must assume that it is not.
268 	 */
269 	if ((get_hwenv() & HW_XEN_HVM) == 0 &&
270 	    is_x86_feature(x86_featureset, X86FSET_TSCP))
271 		patch_tsc_read(TSC_TSCP);
272 	else if (cpuid_getvendor(CPU) == X86_VENDOR_AMD &&
273 	    cpuid_getfamily(CPU) <= 0xf &&
274 	    is_x86_feature(x86_featureset, X86FSET_SSE2))
275 		patch_tsc_read(TSC_RDTSC_MFENCE);
276 	else if (cpuid_getvendor(CPU) == X86_VENDOR_Intel &&
277 	    cpuid_getfamily(CPU) <= 6 &&
278 	    is_x86_feature(x86_featureset, X86FSET_SSE2))
279 		patch_tsc_read(TSC_RDTSC_LFENCE);
280 
281 #endif	/* !__xpv */
282 
283 #if defined(__i386) && !defined(__xpv)
284 	/*
285 	 * Some i386 processors do not implement the rdtsc instruction,
286 	 * or at least they do not implement it correctly. Patch them to
287 	 * return 0.
288 	 */
289 	if (!is_x86_feature(x86_featureset, X86FSET_TSC))
290 		patch_tsc_read(TSC_NONE);
291 #endif	/* __i386 && !__xpv */
292 
293 #if defined(__amd64) && !defined(__xpv)
294 	patch_memops(cpuid_getvendor(CPU));
295 #endif	/* __amd64 && !__xpv */
296 
297 #if !defined(__xpv)
298 	/* XXPV	what, if anything, should be dorked with here under xen? */
299 
300 	/*
301 	 * While we're thinking about the TSC, let's set up %cr4 so that
302 	 * userland can issue rdtsc, and initialize the TSC_AUX value
303 	 * (the cpuid) for the rdtscp instruction on appropriately
304 	 * capable hardware.
305 	 */
306 	if (is_x86_feature(x86_featureset, X86FSET_TSC))
307 		setcr4(getcr4() & ~CR4_TSD);
308 
309 	if (is_x86_feature(x86_featureset, X86FSET_TSCP))
310 		(void) wrmsr(MSR_AMD_TSCAUX, 0);
311 
312 	/*
313 	 * Let's get the other %cr4 stuff while we're here. Note, we defer
314 	 * enabling CR4_SMAP until startup_end(); however, that's importantly
315 	 * before we start other CPUs. That ensures that it will be synced out
316 	 * to other CPUs.
317 	 */
318 	if (is_x86_feature(x86_featureset, X86FSET_DE))
319 		setcr4(getcr4() | CR4_DE);
320 
321 	if (is_x86_feature(x86_featureset, X86FSET_SMEP))
322 		setcr4(getcr4() | CR4_SMEP);
323 #endif /* __xpv */
324 
325 	/*
326 	 * initialize t0
327 	 */
328 	t0.t_stk = (caddr_t)rp - MINFRAME;
329 	t0.t_stkbase = t0stack;
330 	t0.t_pri = maxclsyspri - 3;
331 	t0.t_schedflag = TS_LOAD | TS_DONT_SWAP;
332 	t0.t_procp = &p0;
333 	t0.t_plockp = &p0lock.pl_lock;
334 	t0.t_lwp = &lwp0;
335 	t0.t_forw = &t0;
336 	t0.t_back = &t0;
337 	t0.t_next = &t0;
338 	t0.t_prev = &t0;
339 	t0.t_cpu = cpu[0];
340 	t0.t_disp_queue = &cpu0_disp;
341 	t0.t_bind_cpu = PBIND_NONE;
342 	t0.t_bind_pset = PS_NONE;
343 	t0.t_bindflag = (uchar_t)default_binding_mode;
344 	t0.t_cpupart = &cp_default;
345 	t0.t_clfuncs = &sys_classfuncs.thread;
346 	t0.t_copyops = NULL;
347 	THREAD_ONPROC(&t0, CPU);
348 
349 	lwp0.lwp_thread = &t0;
350 	lwp0.lwp_regs = (void *)rp;
351 	lwp0.lwp_procp = &p0;
352 	t0.t_tid = p0.p_lwpcnt = p0.p_lwprcnt = p0.p_lwpid = 1;
353 
354 	p0.p_exec = NULL;
355 	p0.p_stat = SRUN;
356 	p0.p_flag = SSYS;
357 	p0.p_tlist = &t0;
358 	p0.p_stksize = 2*PAGESIZE;
359 	p0.p_stkpageszc = 0;
360 	p0.p_as = &kas;
361 	p0.p_lockp = &p0lock;
362 	p0.p_brkpageszc = 0;
363 	p0.p_t1_lgrpid = LGRP_NONE;
364 	p0.p_tr_lgrpid = LGRP_NONE;
365 	psecflags_default(&p0.p_secflags);
366 
367 	sigorset(&p0.p_ignore, &ignoredefault);
368 
369 	CPU->cpu_thread = &t0;
370 	bzero(&cpu0_disp, sizeof (disp_t));
371 	CPU->cpu_disp = &cpu0_disp;
372 	CPU->cpu_disp->disp_cpu = CPU;
373 	CPU->cpu_dispthread = &t0;
374 	CPU->cpu_idle_thread = &t0;
375 	CPU->cpu_flags = CPU_READY | CPU_RUNNING | CPU_EXISTS | CPU_ENABLE;
376 	CPU->cpu_dispatch_pri = t0.t_pri;
377 
378 	CPU->cpu_id = 0;
379 
380 	CPU->cpu_pri = 12;		/* initial PIL for the boot CPU */
381 
382 	/*
383 	 * Initialize thread/cpu microstate accounting
384 	 */
385 	init_mstate(&t0, LMS_SYSTEM);
386 	init_cpu_mstate(CPU, CMS_SYSTEM);
387 
388 	/*
389 	 * Initialize lists of available and active CPUs.
390 	 */
391 	cpu_list_init(CPU);
392 
393 	pg_cpu_bootstrap(CPU);
394 
395 	/*
396 	 * Now that we have taken over the GDT, IDT and have initialized
397 	 * active CPU list it's time to inform kmdb if present.
398 	 */
399 	if (boothowto & RB_DEBUG)
400 		kdi_idt_sync();
401 
402 	if (BOP_GETPROPLEN(bootops, "efi-systab") < 0) {
403 		/*
404 		 * In BIOS system, explicitly set console to text mode (0x3)
405 		 * if this is a boot post Fast Reboot, and the console is set
406 		 * to CONS_SCREEN_TEXT.
407 		 */
408 		if (post_fastreboot &&
409 		    boot_console_type(NULL) == CONS_SCREEN_TEXT) {
410 			set_console_mode(0x3);
411 		}
412 	}
413 
414 	/*
415 	 * If requested (boot -d) drop into kmdb.
416 	 *
417 	 * This must be done after cpu_list_init() on the 64-bit kernel
418 	 * since taking a trap requires that we re-compute gsbase based
419 	 * on the cpu list.
420 	 */
421 	if (boothowto & RB_DEBUGENTER)
422 		kmdb_enter();
423 
424 	cpu_vm_data_init(CPU);
425 
426 	rp->r_fp = 0;	/* terminate kernel stack traces! */
427 
428 	prom_init("kernel", (void *)NULL);
429 
430 	/* User-set option overrides firmware value. */
431 	if (bootprop_getval(PLAT_DR_OPTIONS_NAME, &prop_value) == 0) {
432 		plat_dr_options = (uint64_t)prop_value;
433 	}
434 #if defined(__xpv)
435 	/* No support of DR operations on xpv */
436 	plat_dr_options = 0;
437 #else	/* __xpv */
438 	/* Flag PLAT_DR_FEATURE_ENABLED should only be set by DR driver. */
439 	plat_dr_options &= ~PLAT_DR_FEATURE_ENABLED;
440 #ifndef	__amd64
441 	/* Only enable CPU/memory DR on 64 bits kernel. */
442 	plat_dr_options &= ~PLAT_DR_FEATURE_MEMORY;
443 	plat_dr_options &= ~PLAT_DR_FEATURE_CPU;
444 #endif	/* __amd64 */
445 #endif	/* __xpv */
446 
447 	/*
448 	 * Get value of "plat_dr_physmax" boot option.
449 	 * It overrides values calculated from MSCT or SRAT table.
450 	 */
451 	if (bootprop_getval(PLAT_DR_PHYSMAX_NAME, &prop_value) == 0) {
452 		plat_dr_physmax = ((uint64_t)prop_value) >> PAGESHIFT;
453 	}
454 
455 	/* Get value of boot_ncpus. */
456 	if (bootprop_getval(BOOT_NCPUS_NAME, &prop_value) != 0) {
457 		boot_ncpus = NCPU;
458 	} else {
459 		boot_ncpus = (int)prop_value;
460 		if (boot_ncpus <= 0 || boot_ncpus > NCPU)
461 			boot_ncpus = NCPU;
462 	}
463 
464 	/*
465 	 * Set max_ncpus and boot_max_ncpus to boot_ncpus if platform doesn't
466 	 * support CPU DR operations.
467 	 */
468 	if (plat_dr_support_cpu() == 0) {
469 		max_ncpus = boot_max_ncpus = boot_ncpus;
470 	} else {
471 		if (bootprop_getval(PLAT_MAX_NCPUS_NAME, &prop_value) != 0) {
472 			max_ncpus = NCPU;
473 		} else {
474 			max_ncpus = (int)prop_value;
475 			if (max_ncpus <= 0 || max_ncpus > NCPU) {
476 				max_ncpus = NCPU;
477 			}
478 			if (boot_ncpus > max_ncpus) {
479 				boot_ncpus = max_ncpus;
480 			}
481 		}
482 
483 		if (bootprop_getval(BOOT_MAX_NCPUS_NAME, &prop_value) != 0) {
484 			boot_max_ncpus = boot_ncpus;
485 		} else {
486 			boot_max_ncpus = (int)prop_value;
487 			if (boot_max_ncpus <= 0 || boot_max_ncpus > NCPU) {
488 				boot_max_ncpus = boot_ncpus;
489 			} else if (boot_max_ncpus > max_ncpus) {
490 				boot_max_ncpus = max_ncpus;
491 			}
492 		}
493 	}
494 
495 	/*
496 	 * Initialize the lgrp framework
497 	 */
498 	lgrp_init(LGRP_INIT_STAGE1);
499 
500 	if (boothowto & RB_HALT) {
501 		prom_printf("unix: kernel halted by -h flag\n");
502 		prom_enter_mon();
503 	}
504 
505 	ASSERT_STACK_ALIGNED();
506 
507 	/*
508 	 * Fill out cpu_ucode_info.  Update microcode if necessary.
509 	 */
510 	ucode_check(CPU);
511 	cpuid_pass_ucode(CPU, x86_featureset);
512 
513 	if (workaround_errata(CPU) != 0)
514 		panic("critical workaround(s) missing for boot cpu");
515 }
516 
517 
518 void
519 mach_modpath(char *path, const char *filename)
520 {
521 	/*
522 	 * Construct the directory path from the filename.
523 	 */
524 
525 	int len;
526 	char *p;
527 	const char isastr[] = "/amd64";
528 	size_t isalen = strlen(isastr);
529 
530 	len = strlen(SYSTEM_BOOT_PATH "/kernel");
531 	(void) strcpy(path, SYSTEM_BOOT_PATH "/kernel ");
532 	path += len + 1;
533 
534 	if ((p = strrchr(filename, '/')) == NULL)
535 		return;
536 
537 	while (p > filename && *(p - 1) == '/')
538 		p--;	/* remove trailing '/' characters */
539 	if (p == filename)
540 		p++;	/* so "/" -is- the modpath in this case */
541 
542 	/*
543 	 * Remove optional isa-dependent directory name - the module
544 	 * subsystem will put this back again (!)
545 	 */
546 	len = p - filename;
547 	if (len > isalen &&
548 	    strncmp(&filename[len - isalen], isastr, isalen) == 0)
549 		p -= isalen;
550 
551 	/*
552 	 * "/platform/mumblefrotz" + " " + MOD_DEFPATH
553 	 */
554 	len += (p - filename) + 1 + strlen(MOD_DEFPATH) + 1;
555 	(void) strncpy(path, filename, p - filename);
556 }
557