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 2016 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 /*
152 * Initialize idt0, gdt0, ldt0_default, ktss0 and dftss.
153 */
154 init_desctbls();
155
156 /*
157 * lgrp_init() and possibly cpuid_pass1() need PCI config
158 * space access
159 */
160 #if defined(__xpv)
161 if (DOMAIN_IS_INITDOMAIN(xen_info))
162 pci_cfgspace_init();
163 #else
164 pci_cfgspace_init();
165 /*
166 * Initialize the platform type from CPU 0 to ensure that
167 * determine_platform() is only ever called once.
168 */
169 determine_platform();
170 #endif
171
172 /*
173 * The first lightweight pass (pass0) through the cpuid data
174 * was done in locore before mlsetup was called. Do the next
175 * pass in C code.
176 *
177 * The x86_featureset is initialized here based on the capabilities
178 * of the boot CPU. Note that if we choose to support CPUs that have
179 * different feature sets (at which point we would almost certainly
180 * want to set the feature bits to correspond to the feature
181 * minimum) this value may be altered.
182 */
183 cpuid_pass1(cpu[0], x86_featureset);
184
185 #if !defined(__xpv)
186 if ((get_hwenv() & HW_XEN_HVM) != 0)
187 xen_hvm_init();
188
189 /*
190 * Before we do anything with the TSCs, we need to work around
191 * Intel erratum BT81. On some CPUs, warm reset does not
192 * clear the TSC. If we are on such a CPU, we will clear TSC ourselves
193 * here. Other CPUs will clear it when we boot them later, and the
194 * resulting skew will be handled by tsc_sync_master()/_slave();
195 * note that such skew already exists and has to be handled anyway.
196 *
197 * We do this only on metal. This same problem can occur with a
198 * hypervisor that does not happen to virtualise a TSC that starts from
199 * zero, regardless of CPU type; however, we do not expect hypervisors
200 * that do not virtualise TSC that way to handle writes to TSC
201 * correctly, either.
202 */
203 if (get_hwenv() == HW_NATIVE &&
204 cpuid_getvendor(CPU) == X86_VENDOR_Intel &&
205 cpuid_getfamily(CPU) == 6 &&
206 (cpuid_getmodel(CPU) == 0x2d || cpuid_getmodel(CPU) == 0x3e) &&
207 is_x86_feature(x86_featureset, X86FSET_TSC)) {
208 (void) wrmsr(REG_TSC, 0UL);
209 }
210
211 /*
212 * Patch the tsc_read routine with appropriate set of instructions,
213 * depending on the processor family and architecure, to read the
214 * time-stamp counter while ensuring no out-of-order execution.
215 * Patch it while the kernel text is still writable.
216 *
217 * Note: tsc_read is not patched for intel processors whose family
218 * is >6 and for amd whose family >f (in case they don't support rdtscp
219 * instruction, unlikely). By default tsc_read will use cpuid for
220 * serialization in such cases. The following code needs to be
221 * revisited if intel processors of family >= f retains the
222 * instruction serialization nature of mfence instruction.
223 * Note: tsc_read is not patched for x86 processors which do
224 * not support "mfence". By default tsc_read will use cpuid for
225 * serialization in such cases.
226 *
227 * The Xen hypervisor does not correctly report whether rdtscp is
228 * supported or not, so we must assume that it is not.
229 */
230 if ((get_hwenv() & HW_XEN_HVM) == 0 &&
231 is_x86_feature(x86_featureset, X86FSET_TSCP))
232 patch_tsc_read(TSC_TSCP);
233 else if (cpuid_getvendor(CPU) == X86_VENDOR_AMD &&
234 cpuid_getfamily(CPU) <= 0xf &&
235 is_x86_feature(x86_featureset, X86FSET_SSE2))
236 patch_tsc_read(TSC_RDTSC_MFENCE);
237 else if (cpuid_getvendor(CPU) == X86_VENDOR_Intel &&
238 cpuid_getfamily(CPU) <= 6 &&
239 is_x86_feature(x86_featureset, X86FSET_SSE2))
240 patch_tsc_read(TSC_RDTSC_LFENCE);
241
242 #endif /* !__xpv */
243
244 #if defined(__i386) && !defined(__xpv)
245 /*
246 * Some i386 processors do not implement the rdtsc instruction,
247 * or at least they do not implement it correctly. Patch them to
248 * return 0.
249 */
250 if (!is_x86_feature(x86_featureset, X86FSET_TSC))
251 patch_tsc_read(TSC_NONE);
252 #endif /* __i386 && !__xpv */
253
254 #if defined(__amd64) && !defined(__xpv)
255 patch_memops(cpuid_getvendor(CPU));
256 #endif /* __amd64 && !__xpv */
257
258 #if !defined(__xpv)
259 /* XXPV what, if anything, should be dorked with here under xen? */
260
261 /*
262 * While we're thinking about the TSC, let's set up %cr4 so that
263 * userland can issue rdtsc, and initialize the TSC_AUX value
264 * (the cpuid) for the rdtscp instruction on appropriately
265 * capable hardware.
266 */
267 if (is_x86_feature(x86_featureset, X86FSET_TSC))
268 setcr4(getcr4() & ~CR4_TSD);
269
270 if (is_x86_feature(x86_featureset, X86FSET_TSCP))
271 (void) wrmsr(MSR_AMD_TSCAUX, 0);
272
273 /*
274 * Let's get the other %cr4 stuff while we're here. Note, we defer
275 * enabling CR4_SMAP until startup_end(); however, that's importantly
276 * before we start other CPUs. That ensures that it will be synced out
277 * to other CPUs.
278 */
279 if (is_x86_feature(x86_featureset, X86FSET_DE))
280 setcr4(getcr4() | CR4_DE);
281
282 if (is_x86_feature(x86_featureset, X86FSET_SMEP))
283 setcr4(getcr4() | CR4_SMEP);
284 #endif /* __xpv */
285
286 /*
287 * initialize t0
288 */
289 t0.t_stk = (caddr_t)rp - MINFRAME;
290 t0.t_stkbase = t0stack;
291 t0.t_pri = maxclsyspri - 3;
292 t0.t_schedflag = TS_LOAD | TS_DONT_SWAP;
293 t0.t_procp = &p0;
294 t0.t_plockp = &p0lock.pl_lock;
295 t0.t_lwp = &lwp0;
296 t0.t_forw = &t0;
297 t0.t_back = &t0;
298 t0.t_next = &t0;
299 t0.t_prev = &t0;
300 t0.t_cpu = cpu[0];
301 t0.t_disp_queue = &cpu0_disp;
302 t0.t_bind_cpu = PBIND_NONE;
303 t0.t_bind_pset = PS_NONE;
304 t0.t_bindflag = (uchar_t)default_binding_mode;
305 t0.t_cpupart = &cp_default;
306 t0.t_clfuncs = &sys_classfuncs.thread;
307 t0.t_copyops = NULL;
308 THREAD_ONPROC(&t0, CPU);
309
310 lwp0.lwp_thread = &t0;
311 lwp0.lwp_regs = (void *)rp;
312 lwp0.lwp_procp = &p0;
313 t0.t_tid = p0.p_lwpcnt = p0.p_lwprcnt = p0.p_lwpid = 1;
314
315 p0.p_exec = NULL;
316 p0.p_stat = SRUN;
317 p0.p_flag = SSYS;
318 p0.p_tlist = &t0;
319 p0.p_stksize = 2*PAGESIZE;
320 p0.p_stkpageszc = 0;
321 p0.p_as = &kas;
322 p0.p_lockp = &p0lock;
323 p0.p_brkpageszc = 0;
324 p0.p_t1_lgrpid = LGRP_NONE;
325 p0.p_tr_lgrpid = LGRP_NONE;
326 psecflags_default(&p0.p_secflags);
327
328 sigorset(&p0.p_ignore, &ignoredefault);
329
330 CPU->cpu_thread = &t0;
331 bzero(&cpu0_disp, sizeof (disp_t));
332 CPU->cpu_disp = &cpu0_disp;
333 CPU->cpu_disp->disp_cpu = CPU;
334 CPU->cpu_dispthread = &t0;
335 CPU->cpu_idle_thread = &t0;
336 CPU->cpu_flags = CPU_READY | CPU_RUNNING | CPU_EXISTS | CPU_ENABLE;
337 CPU->cpu_dispatch_pri = t0.t_pri;
338
339 CPU->cpu_id = 0;
340
341 CPU->cpu_pri = 12; /* initial PIL for the boot CPU */
342
343 /*
344 * The kernel doesn't use LDTs unless a process explicitly requests one.
345 */
346 p0.p_ldt_desc = null_sdesc;
347
348 /*
349 * Initialize thread/cpu microstate accounting
350 */
351 init_mstate(&t0, LMS_SYSTEM);
352 init_cpu_mstate(CPU, CMS_SYSTEM);
353
354 /*
355 * Initialize lists of available and active CPUs.
356 */
357 cpu_list_init(CPU);
358
359 pg_cpu_bootstrap(CPU);
360
361 /*
362 * Now that we have taken over the GDT, IDT and have initialized
363 * active CPU list it's time to inform kmdb if present.
364 */
365 if (boothowto & RB_DEBUG)
366 kdi_idt_sync();
367
368 if (BOP_GETPROPLEN(bootops, "efi-systab") < 0) {
369 /*
370 * In BIOS system, explicitly set console to text mode (0x3)
371 * if this is a boot post Fast Reboot, and the console is set
372 * to CONS_SCREEN_TEXT.
373 */
374 if (post_fastreboot &&
375 boot_console_type(NULL) == CONS_SCREEN_TEXT) {
376 set_console_mode(0x3);
377 }
378 }
379
380 /*
381 * If requested (boot -d) drop into kmdb.
382 *
383 * This must be done after cpu_list_init() on the 64-bit kernel
384 * since taking a trap requires that we re-compute gsbase based
385 * on the cpu list.
386 */
387 if (boothowto & RB_DEBUGENTER)
388 kmdb_enter();
389
390 cpu_vm_data_init(CPU);
391
392 rp->r_fp = 0; /* terminate kernel stack traces! */
393
394 prom_init("kernel", (void *)NULL);
395
396 /* User-set option overrides firmware value. */
397 if (bootprop_getval(PLAT_DR_OPTIONS_NAME, &prop_value) == 0) {
398 plat_dr_options = (uint64_t)prop_value;
399 }
400 #if defined(__xpv)
401 /* No support of DR operations on xpv */
402 plat_dr_options = 0;
403 #else /* __xpv */
404 /* Flag PLAT_DR_FEATURE_ENABLED should only be set by DR driver. */
405 plat_dr_options &= ~PLAT_DR_FEATURE_ENABLED;
406 #ifndef __amd64
407 /* Only enable CPU/memory DR on 64 bits kernel. */
408 plat_dr_options &= ~PLAT_DR_FEATURE_MEMORY;
409 plat_dr_options &= ~PLAT_DR_FEATURE_CPU;
410 #endif /* __amd64 */
411 #endif /* __xpv */
412
413 /*
414 * Get value of "plat_dr_physmax" boot option.
415 * It overrides values calculated from MSCT or SRAT table.
416 */
417 if (bootprop_getval(PLAT_DR_PHYSMAX_NAME, &prop_value) == 0) {
418 plat_dr_physmax = ((uint64_t)prop_value) >> PAGESHIFT;
419 }
420
421 /* Get value of boot_ncpus. */
422 if (bootprop_getval(BOOT_NCPUS_NAME, &prop_value) != 0) {
423 boot_ncpus = NCPU;
424 } else {
425 boot_ncpus = (int)prop_value;
426 if (boot_ncpus <= 0 || boot_ncpus > NCPU)
427 boot_ncpus = NCPU;
428 }
429
430 /*
431 * Set max_ncpus and boot_max_ncpus to boot_ncpus if platform doesn't
432 * support CPU DR operations.
433 */
434 if (plat_dr_support_cpu() == 0) {
435 max_ncpus = boot_max_ncpus = boot_ncpus;
436 } else {
437 if (bootprop_getval(PLAT_MAX_NCPUS_NAME, &prop_value) != 0) {
438 max_ncpus = NCPU;
439 } else {
440 max_ncpus = (int)prop_value;
441 if (max_ncpus <= 0 || max_ncpus > NCPU) {
442 max_ncpus = NCPU;
443 }
444 if (boot_ncpus > max_ncpus) {
445 boot_ncpus = max_ncpus;
446 }
447 }
448
449 if (bootprop_getval(BOOT_MAX_NCPUS_NAME, &prop_value) != 0) {
450 boot_max_ncpus = boot_ncpus;
451 } else {
452 boot_max_ncpus = (int)prop_value;
453 if (boot_max_ncpus <= 0 || boot_max_ncpus > NCPU) {
454 boot_max_ncpus = boot_ncpus;
455 } else if (boot_max_ncpus > max_ncpus) {
456 boot_max_ncpus = max_ncpus;
457 }
458 }
459 }
460
461 /*
462 * Initialize the lgrp framework
463 */
464 lgrp_init(LGRP_INIT_STAGE1);
465
466 if (boothowto & RB_HALT) {
467 prom_printf("unix: kernel halted by -h flag\n");
468 prom_enter_mon();
469 }
470
471 ASSERT_STACK_ALIGNED();
472
473 /*
474 * Fill out cpu_ucode_info. Update microcode if necessary.
475 */
476 ucode_check(CPU);
477
478 if (workaround_errata(CPU) != 0)
479 panic("critical workaround(s) missing for boot cpu");
480 }
481
482
483 void
484 mach_modpath(char *path, const char *filename)
485 {
486 /*
487 * Construct the directory path from the filename.
488 */
489
490 int len;
491 char *p;
492 const char isastr[] = "/amd64";
493 size_t isalen = strlen(isastr);
494
495 len = strlen(SYSTEM_BOOT_PATH "/kernel");
496 (void) strcpy(path, SYSTEM_BOOT_PATH "/kernel ");
497 path += len + 1;
498
499 if ((p = strrchr(filename, '/')) == NULL)
500 return;
501
502 while (p > filename && *(p - 1) == '/')
503 p--; /* remove trailing '/' characters */
504 if (p == filename)
505 p++; /* so "/" -is- the modpath in this case */
506
507 /*
508 * Remove optional isa-dependent directory name - the module
509 * subsystem will put this back again (!)
510 */
511 len = p - filename;
512 if (len > isalen &&
513 strncmp(&filename[len - isalen], isastr, isalen) == 0)
514 p -= isalen;
515
516 /*
517 * "/platform/mumblefrotz" + " " + MOD_DEFPATH
518 */
519 len += (p - filename) + 1 + strlen(MOD_DEFPATH) + 1;
520 (void) strncpy(path, filename, p - filename);
521 }