xref: /illumos-gate/usr/src/uts/i86pc/os/mlsetup.c (revision 5bb0bdfe588c5df0f63ff8ac292cd608a5f4492a)
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 2018 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/bootvfs.h>
65 #include <sys/tsc.h>
66 #ifdef __xpv
67 #include <sys/hypervisor.h>
68 #else
69 #include <sys/xpv_support.h>
70 #endif
71 
72 /*
73  * some globals for patching the result of cpuid
74  * to solve problems w/ creative cpu vendors
75  */
76 
77 extern uint32_t cpuid_feature_ecx_include;
78 extern uint32_t cpuid_feature_ecx_exclude;
79 extern uint32_t cpuid_feature_edx_include;
80 extern uint32_t cpuid_feature_edx_exclude;
81 
82 /*
83  * Set console mode
84  */
85 static void
86 set_console_mode(uint8_t val)
87 {
88 	struct bop_regs rp = {0};
89 
90 	rp.eax.byte.ah = 0x0;
91 	rp.eax.byte.al = val;
92 	rp.ebx.word.bx = 0x0;
93 
94 	BOP_DOINT(bootops, 0x10, &rp);
95 }
96 
97 
98 /*
99  * Setup routine called right before main(). Interposing this function
100  * before main() allows us to call it in a machine-independent fashion.
101  */
102 void
103 mlsetup(struct regs *rp)
104 {
105 	u_longlong_t prop_value;
106 	extern struct classfuncs sys_classfuncs;
107 	extern disp_t cpu0_disp;
108 	extern char t0stack[];
109 	extern int post_fastreboot;
110 	extern uint64_t plat_dr_options;
111 
112 	ASSERT_STACK_ALIGNED();
113 
114 	/*
115 	 * initialize cpu_self
116 	 */
117 	cpu[0]->cpu_self = cpu[0];
118 
119 #if defined(__xpv)
120 	/*
121 	 * Point at the hypervisor's virtual cpu structure
122 	 */
123 	cpu[0]->cpu_m.mcpu_vcpu_info = &HYPERVISOR_shared_info->vcpu_info[0];
124 #endif
125 
126 	/*
127 	 * check if we've got special bits to clear or set
128 	 * when checking cpu features
129 	 */
130 
131 	if (bootprop_getval("cpuid_feature_ecx_include", &prop_value) != 0)
132 		cpuid_feature_ecx_include = 0;
133 	else
134 		cpuid_feature_ecx_include = (uint32_t)prop_value;
135 
136 	if (bootprop_getval("cpuid_feature_ecx_exclude", &prop_value) != 0)
137 		cpuid_feature_ecx_exclude = 0;
138 	else
139 		cpuid_feature_ecx_exclude = (uint32_t)prop_value;
140 
141 	if (bootprop_getval("cpuid_feature_edx_include", &prop_value) != 0)
142 		cpuid_feature_edx_include = 0;
143 	else
144 		cpuid_feature_edx_include = (uint32_t)prop_value;
145 
146 	if (bootprop_getval("cpuid_feature_edx_exclude", &prop_value) != 0)
147 		cpuid_feature_edx_exclude = 0;
148 	else
149 		cpuid_feature_edx_exclude = (uint32_t)prop_value;
150 
151 #if !defined(__xpv)
152 	/*
153 	 * Check to see if KPTI has been explicitly enabled or disabled.
154 	 * We have to check this before init_desctbls().
155 	 */
156 	if (bootprop_getval("kpti", &prop_value) == 0) {
157 		kpti_enable = (uint64_t)(prop_value == 1);
158 		prom_printf("unix: forcing kpti to %s due to boot argument\n",
159 		    (kpti_enable == 1) ? "ON" : "OFF");
160 	} else {
161 		kpti_enable = 1;
162 	}
163 
164 	if (bootprop_getval("pcid", &prop_value) == 0 && prop_value == 0) {
165 		prom_printf("unix: forcing pcid to OFF due to boot argument\n");
166 		x86_use_pcid = 0;
167 	} else if (kpti_enable != 1) {
168 		x86_use_pcid = 0;
169 	}
170 #endif
171 
172 	/*
173 	 * Initialize idt0, gdt0, ldt0_default, ktss0 and dftss.
174 	 */
175 	init_desctbls();
176 
177 	/*
178 	 * lgrp_init() and possibly cpuid_pass1() need PCI config
179 	 * space access
180 	 */
181 #if defined(__xpv)
182 	if (DOMAIN_IS_INITDOMAIN(xen_info))
183 		pci_cfgspace_init();
184 #else
185 	pci_cfgspace_init();
186 	/*
187 	 * Initialize the platform type from CPU 0 to ensure that
188 	 * determine_platform() is only ever called once.
189 	 */
190 	determine_platform();
191 #endif
192 
193 	/*
194 	 * The first lightweight pass (pass0) through the cpuid data
195 	 * was done in locore before mlsetup was called.  Do the next
196 	 * pass in C code.
197 	 *
198 	 * The x86_featureset is initialized here based on the capabilities
199 	 * of the boot CPU.  Note that if we choose to support CPUs that have
200 	 * different feature sets (at which point we would almost certainly
201 	 * want to set the feature bits to correspond to the feature
202 	 * minimum) this value may be altered.
203 	 */
204 	cpuid_pass1(cpu[0], x86_featureset);
205 
206 #if !defined(__xpv)
207 	if ((get_hwenv() & HW_XEN_HVM) != 0)
208 		xen_hvm_init();
209 
210 	/*
211 	 * Before we do anything with the TSCs, we need to work around
212 	 * Intel erratum BT81.  On some CPUs, warm reset does not
213 	 * clear the TSC.  If we are on such a CPU, we will clear TSC ourselves
214 	 * here.  Other CPUs will clear it when we boot them later, and the
215 	 * resulting skew will be handled by tsc_sync_master()/_slave();
216 	 * note that such skew already exists and has to be handled anyway.
217 	 *
218 	 * We do this only on metal.  This same problem can occur with a
219 	 * hypervisor that does not happen to virtualise a TSC that starts from
220 	 * zero, regardless of CPU type; however, we do not expect hypervisors
221 	 * that do not virtualise TSC that way to handle writes to TSC
222 	 * correctly, either.
223 	 */
224 	if (get_hwenv() == HW_NATIVE &&
225 	    cpuid_getvendor(CPU) == X86_VENDOR_Intel &&
226 	    cpuid_getfamily(CPU) == 6 &&
227 	    (cpuid_getmodel(CPU) == 0x2d || cpuid_getmodel(CPU) == 0x3e) &&
228 	    is_x86_feature(x86_featureset, X86FSET_TSC)) {
229 		(void) wrmsr(REG_TSC, 0UL);
230 	}
231 
232 	/*
233 	 * Patch the tsc_read routine with appropriate set of instructions,
234 	 * depending on the processor family and architecure, to read the
235 	 * time-stamp counter while ensuring no out-of-order execution.
236 	 * Patch it while the kernel text is still writable.
237 	 *
238 	 * Note: tsc_read is not patched for intel processors whose family
239 	 * is >6 and for amd whose family >f (in case they don't support rdtscp
240 	 * instruction, unlikely). By default tsc_read will use cpuid for
241 	 * serialization in such cases. The following code needs to be
242 	 * revisited if intel processors of family >= f retains the
243 	 * instruction serialization nature of mfence instruction.
244 	 * Note: tsc_read is not patched for x86 processors which do
245 	 * not support "mfence". By default tsc_read will use cpuid for
246 	 * serialization in such cases.
247 	 *
248 	 * The Xen hypervisor does not correctly report whether rdtscp is
249 	 * supported or not, so we must assume that it is not.
250 	 */
251 	if ((get_hwenv() & HW_XEN_HVM) == 0 &&
252 	    is_x86_feature(x86_featureset, X86FSET_TSCP))
253 		patch_tsc_read(TSC_TSCP);
254 	else if (cpuid_getvendor(CPU) == X86_VENDOR_AMD &&
255 	    cpuid_getfamily(CPU) <= 0xf &&
256 	    is_x86_feature(x86_featureset, X86FSET_SSE2))
257 		patch_tsc_read(TSC_RDTSC_MFENCE);
258 	else if (cpuid_getvendor(CPU) == X86_VENDOR_Intel &&
259 	    cpuid_getfamily(CPU) <= 6 &&
260 	    is_x86_feature(x86_featureset, X86FSET_SSE2))
261 		patch_tsc_read(TSC_RDTSC_LFENCE);
262 
263 #endif	/* !__xpv */
264 
265 #if defined(__i386) && !defined(__xpv)
266 	/*
267 	 * Some i386 processors do not implement the rdtsc instruction,
268 	 * or at least they do not implement it correctly. Patch them to
269 	 * return 0.
270 	 */
271 	if (!is_x86_feature(x86_featureset, X86FSET_TSC))
272 		patch_tsc_read(TSC_NONE);
273 #endif	/* __i386 && !__xpv */
274 
275 #if defined(__amd64) && !defined(__xpv)
276 	patch_memops(cpuid_getvendor(CPU));
277 #endif	/* __amd64 && !__xpv */
278 
279 #if !defined(__xpv)
280 	/* XXPV	what, if anything, should be dorked with here under xen? */
281 
282 	/*
283 	 * While we're thinking about the TSC, let's set up %cr4 so that
284 	 * userland can issue rdtsc, and initialize the TSC_AUX value
285 	 * (the cpuid) for the rdtscp instruction on appropriately
286 	 * capable hardware.
287 	 */
288 	if (is_x86_feature(x86_featureset, X86FSET_TSC))
289 		setcr4(getcr4() & ~CR4_TSD);
290 
291 	if (is_x86_feature(x86_featureset, X86FSET_TSCP))
292 		(void) wrmsr(MSR_AMD_TSCAUX, 0);
293 
294 	/*
295 	 * Let's get the other %cr4 stuff while we're here. Note, we defer
296 	 * enabling CR4_SMAP until startup_end(); however, that's importantly
297 	 * before we start other CPUs. That ensures that it will be synced out
298 	 * to other CPUs.
299 	 */
300 	if (is_x86_feature(x86_featureset, X86FSET_DE))
301 		setcr4(getcr4() | CR4_DE);
302 
303 	if (is_x86_feature(x86_featureset, X86FSET_SMEP))
304 		setcr4(getcr4() | CR4_SMEP);
305 #endif /* __xpv */
306 
307 	/*
308 	 * initialize t0
309 	 */
310 	t0.t_stk = (caddr_t)rp - MINFRAME;
311 	t0.t_stkbase = t0stack;
312 	t0.t_pri = maxclsyspri - 3;
313 	t0.t_schedflag = TS_LOAD | TS_DONT_SWAP;
314 	t0.t_procp = &p0;
315 	t0.t_plockp = &p0lock.pl_lock;
316 	t0.t_lwp = &lwp0;
317 	t0.t_forw = &t0;
318 	t0.t_back = &t0;
319 	t0.t_next = &t0;
320 	t0.t_prev = &t0;
321 	t0.t_cpu = cpu[0];
322 	t0.t_disp_queue = &cpu0_disp;
323 	t0.t_bind_cpu = PBIND_NONE;
324 	t0.t_bind_pset = PS_NONE;
325 	t0.t_bindflag = (uchar_t)default_binding_mode;
326 	t0.t_cpupart = &cp_default;
327 	t0.t_clfuncs = &sys_classfuncs.thread;
328 	t0.t_copyops = NULL;
329 	THREAD_ONPROC(&t0, CPU);
330 
331 	lwp0.lwp_thread = &t0;
332 	lwp0.lwp_regs = (void *)rp;
333 	lwp0.lwp_procp = &p0;
334 	t0.t_tid = p0.p_lwpcnt = p0.p_lwprcnt = p0.p_lwpid = 1;
335 
336 	p0.p_exec = NULL;
337 	p0.p_stat = SRUN;
338 	p0.p_flag = SSYS;
339 	p0.p_tlist = &t0;
340 	p0.p_stksize = 2*PAGESIZE;
341 	p0.p_stkpageszc = 0;
342 	p0.p_as = &kas;
343 	p0.p_lockp = &p0lock;
344 	p0.p_brkpageszc = 0;
345 	p0.p_t1_lgrpid = LGRP_NONE;
346 	p0.p_tr_lgrpid = LGRP_NONE;
347 	psecflags_default(&p0.p_secflags);
348 
349 	sigorset(&p0.p_ignore, &ignoredefault);
350 
351 	CPU->cpu_thread = &t0;
352 	bzero(&cpu0_disp, sizeof (disp_t));
353 	CPU->cpu_disp = &cpu0_disp;
354 	CPU->cpu_disp->disp_cpu = CPU;
355 	CPU->cpu_dispthread = &t0;
356 	CPU->cpu_idle_thread = &t0;
357 	CPU->cpu_flags = CPU_READY | CPU_RUNNING | CPU_EXISTS | CPU_ENABLE;
358 	CPU->cpu_dispatch_pri = t0.t_pri;
359 
360 	CPU->cpu_id = 0;
361 
362 	CPU->cpu_pri = 12;		/* initial PIL for the boot CPU */
363 
364 	/*
365 	 * Initialize thread/cpu microstate accounting
366 	 */
367 	init_mstate(&t0, LMS_SYSTEM);
368 	init_cpu_mstate(CPU, CMS_SYSTEM);
369 
370 	/*
371 	 * Initialize lists of available and active CPUs.
372 	 */
373 	cpu_list_init(CPU);
374 
375 	pg_cpu_bootstrap(CPU);
376 
377 	/*
378 	 * Now that we have taken over the GDT, IDT and have initialized
379 	 * active CPU list it's time to inform kmdb if present.
380 	 */
381 	if (boothowto & RB_DEBUG)
382 		kdi_idt_sync();
383 
384 	if (BOP_GETPROPLEN(bootops, "efi-systab") < 0) {
385 		/*
386 		 * In BIOS system, explicitly set console to text mode (0x3)
387 		 * if this is a boot post Fast Reboot, and the console is set
388 		 * to CONS_SCREEN_TEXT.
389 		 */
390 		if (post_fastreboot &&
391 		    boot_console_type(NULL) == CONS_SCREEN_TEXT) {
392 			set_console_mode(0x3);
393 		}
394 	}
395 
396 	/*
397 	 * If requested (boot -d) drop into kmdb.
398 	 *
399 	 * This must be done after cpu_list_init() on the 64-bit kernel
400 	 * since taking a trap requires that we re-compute gsbase based
401 	 * on the cpu list.
402 	 */
403 	if (boothowto & RB_DEBUGENTER)
404 		kmdb_enter();
405 
406 	cpu_vm_data_init(CPU);
407 
408 	rp->r_fp = 0;	/* terminate kernel stack traces! */
409 
410 	prom_init("kernel", (void *)NULL);
411 
412 	/* User-set option overrides firmware value. */
413 	if (bootprop_getval(PLAT_DR_OPTIONS_NAME, &prop_value) == 0) {
414 		plat_dr_options = (uint64_t)prop_value;
415 	}
416 #if defined(__xpv)
417 	/* No support of DR operations on xpv */
418 	plat_dr_options = 0;
419 #else	/* __xpv */
420 	/* Flag PLAT_DR_FEATURE_ENABLED should only be set by DR driver. */
421 	plat_dr_options &= ~PLAT_DR_FEATURE_ENABLED;
422 #ifndef	__amd64
423 	/* Only enable CPU/memory DR on 64 bits kernel. */
424 	plat_dr_options &= ~PLAT_DR_FEATURE_MEMORY;
425 	plat_dr_options &= ~PLAT_DR_FEATURE_CPU;
426 #endif	/* __amd64 */
427 #endif	/* __xpv */
428 
429 	/*
430 	 * Get value of "plat_dr_physmax" boot option.
431 	 * It overrides values calculated from MSCT or SRAT table.
432 	 */
433 	if (bootprop_getval(PLAT_DR_PHYSMAX_NAME, &prop_value) == 0) {
434 		plat_dr_physmax = ((uint64_t)prop_value) >> PAGESHIFT;
435 	}
436 
437 	/* Get value of boot_ncpus. */
438 	if (bootprop_getval(BOOT_NCPUS_NAME, &prop_value) != 0) {
439 		boot_ncpus = NCPU;
440 	} else {
441 		boot_ncpus = (int)prop_value;
442 		if (boot_ncpus <= 0 || boot_ncpus > NCPU)
443 			boot_ncpus = NCPU;
444 	}
445 
446 	/*
447 	 * Set max_ncpus and boot_max_ncpus to boot_ncpus if platform doesn't
448 	 * support CPU DR operations.
449 	 */
450 	if (plat_dr_support_cpu() == 0) {
451 		max_ncpus = boot_max_ncpus = boot_ncpus;
452 	} else {
453 		if (bootprop_getval(PLAT_MAX_NCPUS_NAME, &prop_value) != 0) {
454 			max_ncpus = NCPU;
455 		} else {
456 			max_ncpus = (int)prop_value;
457 			if (max_ncpus <= 0 || max_ncpus > NCPU) {
458 				max_ncpus = NCPU;
459 			}
460 			if (boot_ncpus > max_ncpus) {
461 				boot_ncpus = max_ncpus;
462 			}
463 		}
464 
465 		if (bootprop_getval(BOOT_MAX_NCPUS_NAME, &prop_value) != 0) {
466 			boot_max_ncpus = boot_ncpus;
467 		} else {
468 			boot_max_ncpus = (int)prop_value;
469 			if (boot_max_ncpus <= 0 || boot_max_ncpus > NCPU) {
470 				boot_max_ncpus = boot_ncpus;
471 			} else if (boot_max_ncpus > max_ncpus) {
472 				boot_max_ncpus = max_ncpus;
473 			}
474 		}
475 	}
476 
477 	/*
478 	 * Initialize the lgrp framework
479 	 */
480 	lgrp_init(LGRP_INIT_STAGE1);
481 
482 	if (boothowto & RB_HALT) {
483 		prom_printf("unix: kernel halted by -h flag\n");
484 		prom_enter_mon();
485 	}
486 
487 	ASSERT_STACK_ALIGNED();
488 
489 	/*
490 	 * Fill out cpu_ucode_info.  Update microcode if necessary.
491 	 */
492 	ucode_check(CPU);
493 	cpuid_pass_ucode(CPU, x86_featureset);
494 
495 	if (workaround_errata(CPU) != 0)
496 		panic("critical workaround(s) missing for boot cpu");
497 }
498 
499 
500 void
501 mach_modpath(char *path, const char *filename)
502 {
503 	/*
504 	 * Construct the directory path from the filename.
505 	 */
506 
507 	int len;
508 	char *p;
509 	const char isastr[] = "/amd64";
510 	size_t isalen = strlen(isastr);
511 
512 	len = strlen(SYSTEM_BOOT_PATH "/kernel");
513 	(void) strcpy(path, SYSTEM_BOOT_PATH "/kernel ");
514 	path += len + 1;
515 
516 	if ((p = strrchr(filename, '/')) == NULL)
517 		return;
518 
519 	while (p > filename && *(p - 1) == '/')
520 		p--;	/* remove trailing '/' characters */
521 	if (p == filename)
522 		p++;	/* so "/" -is- the modpath in this case */
523 
524 	/*
525 	 * Remove optional isa-dependent directory name - the module
526 	 * subsystem will put this back again (!)
527 	 */
528 	len = p - filename;
529 	if (len > isalen &&
530 	    strncmp(&filename[len - isalen], isastr, isalen) == 0)
531 		p -= isalen;
532 
533 	/*
534 	 * "/platform/mumblefrotz" + " " + MOD_DEFPATH
535 	 */
536 	len += (p - filename) + 1 + strlen(MOD_DEFPATH) + 1;
537 	(void) strncpy(path, filename, p - filename);
538 }
539