xref: /titanic_41/usr/src/uts/i86pc/os/mlsetup.c (revision d5ace9454616652a717c9831d949dffa319381f9)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/sysmacros.h>
28 #include <sys/disp.h>
29 #include <sys/promif.h>
30 #include <sys/clock.h>
31 #include <sys/cpuvar.h>
32 #include <sys/stack.h>
33 #include <vm/as.h>
34 #include <vm/hat.h>
35 #include <sys/reboot.h>
36 #include <sys/avintr.h>
37 #include <sys/vtrace.h>
38 #include <sys/proc.h>
39 #include <sys/thread.h>
40 #include <sys/cpupart.h>
41 #include <sys/pset.h>
42 #include <sys/copyops.h>
43 #include <sys/pg.h>
44 #include <sys/disp.h>
45 #include <sys/debug.h>
46 #include <sys/sunddi.h>
47 #include <sys/x86_archext.h>
48 #include <sys/privregs.h>
49 #include <sys/machsystm.h>
50 #include <sys/ontrap.h>
51 #include <sys/bootconf.h>
52 #include <sys/kdi_machimpl.h>
53 #include <sys/archsystm.h>
54 #include <sys/promif.h>
55 #include <sys/bootconf.h>
56 #include <sys/pci_cfgspace.h>
57 #ifdef __xpv
58 #include <sys/hypervisor.h>
59 #endif
60 
61 /*
62  * some globals for patching the result of cpuid
63  * to solve problems w/ creative cpu vendors
64  */
65 
66 extern uint32_t cpuid_feature_ecx_include;
67 extern uint32_t cpuid_feature_ecx_exclude;
68 extern uint32_t cpuid_feature_edx_include;
69 extern uint32_t cpuid_feature_edx_exclude;
70 
71 /*
72  * Dummy spl priority masks
73  */
74 static unsigned char dummy_cpu_pri[MAXIPL + 1] = {
75 	0xf, 0xf, 0xf, 0xf, 0xf, 0xf, 0xf, 0xf,
76 	0xf, 0xf, 0xf, 0xf, 0xf, 0xf, 0xf, 0xf, 0xf
77 };
78 
79 
80 /*
81  * Setup routine called right before main(). Interposing this function
82  * before main() allows us to call it in a machine-independent fashion.
83  */
84 void
85 mlsetup(struct regs *rp)
86 {
87 	u_longlong_t prop_value;
88 	extern struct classfuncs sys_classfuncs;
89 	extern disp_t cpu0_disp;
90 	extern char t0stack[];
91 
92 	ASSERT_STACK_ALIGNED();
93 
94 	/*
95 	 * initialize cpu_self
96 	 */
97 	cpu[0]->cpu_self = cpu[0];
98 
99 #if defined(__xpv)
100 	/*
101 	 * Point at the hypervisor's virtual cpu structure
102 	 */
103 	cpu[0]->cpu_m.mcpu_vcpu_info = &HYPERVISOR_shared_info->vcpu_info[0];
104 #endif
105 
106 	/*
107 	 * Set up dummy cpu_pri_data values till psm spl code is
108 	 * installed.  This allows splx() to work on amd64.
109 	 */
110 
111 	cpu[0]->cpu_pri_data = dummy_cpu_pri;
112 
113 	/*
114 	 * check if we've got special bits to clear or set
115 	 * when checking cpu features
116 	 */
117 
118 	if (bootprop_getval("cpuid_feature_ecx_include", &prop_value) != 0)
119 		cpuid_feature_ecx_include = 0;
120 	else
121 		cpuid_feature_ecx_include = (uint32_t)prop_value;
122 
123 	if (bootprop_getval("cpuid_feature_ecx_exclude", &prop_value) != 0)
124 		cpuid_feature_ecx_exclude = 0;
125 	else
126 		cpuid_feature_ecx_exclude = (uint32_t)prop_value;
127 
128 	if (bootprop_getval("cpuid_feature_edx_include", &prop_value) != 0)
129 		cpuid_feature_edx_include = 0;
130 	else
131 		cpuid_feature_edx_include = (uint32_t)prop_value;
132 
133 	if (bootprop_getval("cpuid_feature_edx_exclude", &prop_value) != 0)
134 		cpuid_feature_edx_exclude = 0;
135 	else
136 		cpuid_feature_edx_exclude = (uint32_t)prop_value;
137 
138 	/*
139 	 * The first lightweight pass (pass0) through the cpuid data
140 	 * was done in locore before mlsetup was called.  Do the next
141 	 * pass in C code.
142 	 *
143 	 * The x86_feature bits are set here on the basis of the capabilities
144 	 * of the boot CPU.  Note that if we choose to support CPUs that have
145 	 * different feature sets (at which point we would almost certainly
146 	 * want to set the feature bits to correspond to the feature
147 	 * minimum) this value may be altered.
148 	 */
149 	x86_feature = cpuid_pass1(cpu[0]);
150 
151 	/*
152 	 * Initialize idt0, gdt0, ldt0_default, ktss0 and dftss.
153 	 */
154 	init_desctbls();
155 
156 #if !defined(__xpv)
157 
158 	/*
159 	 * Patch the tsc_read routine with appropriate set of instructions,
160 	 * depending on the processor family and architecure, to read the
161 	 * time-stamp counter while ensuring no out-of-order execution.
162 	 * Patch it while the kernel text is still writable.
163 	 *
164 	 * Note: tsc_read is not patched for intel processors whose family
165 	 * is >6 and for amd whose family >f (in case they don't support rdtscp
166 	 * instruction, unlikely). By default tsc_read will use cpuid for
167 	 * serialization in such cases. The following code needs to be
168 	 * revisited if intel processors of family >= f retains the
169 	 * instruction serialization nature of mfence instruction.
170 	 * Note: tsc_read is not patched for x86 processors which do
171 	 * not support "mfence". By default tsc_read will use cpuid for
172 	 * serialization in such cases.
173 	 *
174 	 * The Xen hypervisor does not correctly report whether rdtscp is
175 	 * supported or not, so we must assume that it is not.
176 	 */
177 	if (get_hwenv() != HW_XEN_HVM && (x86_feature & X86_TSCP))
178 		patch_tsc_read(X86_HAVE_TSCP);
179 	else if (cpuid_getvendor(CPU) == X86_VENDOR_AMD &&
180 	    cpuid_getfamily(CPU) <= 0xf && (x86_feature & X86_SSE2) != 0)
181 		patch_tsc_read(X86_TSC_MFENCE);
182 	else if (cpuid_getvendor(CPU) == X86_VENDOR_Intel &&
183 	    cpuid_getfamily(CPU) <= 6 && (x86_feature & X86_SSE2) != 0)
184 		patch_tsc_read(X86_TSC_LFENCE);
185 
186 #endif	/* !__xpv */
187 
188 #if defined(__i386) && !defined(__xpv)
189 	/*
190 	 * Some i386 processors do not implement the rdtsc instruction,
191 	 * or at least they do not implement it correctly. Patch them to
192 	 * return 0.
193 	 */
194 	if ((x86_feature & X86_TSC) == 0)
195 		patch_tsc_read(X86_NO_TSC);
196 #endif	/* __i386 && !__xpv */
197 
198 #if defined(__amd64) && !defined(__xpv)
199 	patch_memops(cpuid_getvendor(CPU));
200 #endif	/* __amd64 && !__xpv */
201 
202 #if !defined(__xpv)
203 	/* XXPV	what, if anything, should be dorked with here under xen? */
204 
205 	/*
206 	 * While we're thinking about the TSC, let's set up %cr4 so that
207 	 * userland can issue rdtsc, and initialize the TSC_AUX value
208 	 * (the cpuid) for the rdtscp instruction on appropriately
209 	 * capable hardware.
210 	 */
211 	if (x86_feature & X86_TSC)
212 		setcr4(getcr4() & ~CR4_TSD);
213 
214 	if (x86_feature & X86_TSCP)
215 		(void) wrmsr(MSR_AMD_TSCAUX, 0);
216 
217 	if (x86_feature & X86_DE)
218 		setcr4(getcr4() | CR4_DE);
219 #endif /* __xpv */
220 
221 	/*
222 	 * initialize t0
223 	 */
224 	t0.t_stk = (caddr_t)rp - MINFRAME;
225 	t0.t_stkbase = t0stack;
226 	t0.t_pri = maxclsyspri - 3;
227 	t0.t_schedflag = TS_LOAD | TS_DONT_SWAP;
228 	t0.t_procp = &p0;
229 	t0.t_plockp = &p0lock.pl_lock;
230 	t0.t_lwp = &lwp0;
231 	t0.t_forw = &t0;
232 	t0.t_back = &t0;
233 	t0.t_next = &t0;
234 	t0.t_prev = &t0;
235 	t0.t_cpu = cpu[0];
236 	t0.t_disp_queue = &cpu0_disp;
237 	t0.t_bind_cpu = PBIND_NONE;
238 	t0.t_bind_pset = PS_NONE;
239 	t0.t_bindflag = (uchar_t)default_binding_mode;
240 	t0.t_cpupart = &cp_default;
241 	t0.t_clfuncs = &sys_classfuncs.thread;
242 	t0.t_copyops = NULL;
243 	THREAD_ONPROC(&t0, CPU);
244 
245 	lwp0.lwp_thread = &t0;
246 	lwp0.lwp_regs = (void *)rp;
247 	lwp0.lwp_procp = &p0;
248 	t0.t_tid = p0.p_lwpcnt = p0.p_lwprcnt = p0.p_lwpid = 1;
249 
250 	p0.p_exec = NULL;
251 	p0.p_stat = SRUN;
252 	p0.p_flag = SSYS;
253 	p0.p_tlist = &t0;
254 	p0.p_stksize = 2*PAGESIZE;
255 	p0.p_stkpageszc = 0;
256 	p0.p_as = &kas;
257 	p0.p_lockp = &p0lock;
258 	p0.p_brkpageszc = 0;
259 	p0.p_t1_lgrpid = LGRP_NONE;
260 	p0.p_tr_lgrpid = LGRP_NONE;
261 	sigorset(&p0.p_ignore, &ignoredefault);
262 
263 	CPU->cpu_thread = &t0;
264 	bzero(&cpu0_disp, sizeof (disp_t));
265 	CPU->cpu_disp = &cpu0_disp;
266 	CPU->cpu_disp->disp_cpu = CPU;
267 	CPU->cpu_dispthread = &t0;
268 	CPU->cpu_idle_thread = &t0;
269 	CPU->cpu_flags = CPU_READY | CPU_RUNNING | CPU_EXISTS | CPU_ENABLE;
270 	CPU->cpu_dispatch_pri = t0.t_pri;
271 
272 	CPU->cpu_id = 0;
273 
274 	CPU->cpu_pri = 12;		/* initial PIL for the boot CPU */
275 
276 	/*
277 	 * The kernel doesn't use LDTs unless a process explicitly requests one.
278 	 */
279 	p0.p_ldt_desc = null_sdesc;
280 
281 	/*
282 	 * Initialize thread/cpu microstate accounting
283 	 */
284 	init_mstate(&t0, LMS_SYSTEM);
285 	init_cpu_mstate(CPU, CMS_SYSTEM);
286 
287 	/*
288 	 * Initialize lists of available and active CPUs.
289 	 */
290 	cpu_list_init(CPU);
291 
292 	pg_cpu_bootstrap(CPU);
293 
294 	/*
295 	 * Now that we have taken over the GDT, IDT and have initialized
296 	 * active CPU list it's time to inform kmdb if present.
297 	 */
298 	if (boothowto & RB_DEBUG)
299 		kdi_idt_sync();
300 
301 	/*
302 	 * If requested (boot -d) drop into kmdb.
303 	 *
304 	 * This must be done after cpu_list_init() on the 64-bit kernel
305 	 * since taking a trap requires that we re-compute gsbase based
306 	 * on the cpu list.
307 	 */
308 	if (boothowto & RB_DEBUGENTER)
309 		kmdb_enter();
310 
311 	cpu_vm_data_init(CPU);
312 
313 	/* lgrp_init() needs PCI config space access */
314 #if defined(__xpv)
315 	if (DOMAIN_IS_INITDOMAIN(xen_info))
316 		pci_cfgspace_init();
317 #else
318 	pci_cfgspace_init();
319 #endif
320 
321 	rp->r_fp = 0;	/* terminate kernel stack traces! */
322 
323 	prom_init("kernel", (void *)NULL);
324 
325 	if (bootprop_getval("boot-ncpus", &prop_value) != 0)
326 		boot_ncpus = NCPU;
327 	else {
328 		boot_ncpus = (int)prop_value;
329 		if (boot_ncpus <= 0 || boot_ncpus > NCPU)
330 			boot_ncpus = NCPU;
331 	}
332 
333 	max_ncpus = boot_max_ncpus = boot_ncpus;
334 
335 	/*
336 	 * Initialize the lgrp framework
337 	 */
338 	lgrp_init();
339 
340 	if (boothowto & RB_HALT) {
341 		prom_printf("unix: kernel halted by -h flag\n");
342 		prom_enter_mon();
343 	}
344 
345 	ASSERT_STACK_ALIGNED();
346 
347 	/*
348 	 * Fill out cpu_ucode_info.  Update microcode if necessary.
349 	 */
350 	ucode_check(CPU);
351 
352 	if (workaround_errata(CPU) != 0)
353 		panic("critical workaround(s) missing for boot cpu");
354 }
355 
356 
357 void
358 mach_modpath(char *path, const char *filename)
359 {
360 	/*
361 	 * Construct the directory path from the filename.
362 	 */
363 
364 	int len;
365 	char *p;
366 	const char isastr[] = "/amd64";
367 	size_t isalen = strlen(isastr);
368 
369 	if ((p = strrchr(filename, '/')) == NULL)
370 		return;
371 
372 	while (p > filename && *(p - 1) == '/')
373 		p--;	/* remove trailing '/' characters */
374 	if (p == filename)
375 		p++;	/* so "/" -is- the modpath in this case */
376 
377 	/*
378 	 * Remove optional isa-dependent directory name - the module
379 	 * subsystem will put this back again (!)
380 	 */
381 	len = p - filename;
382 	if (len > isalen &&
383 	    strncmp(&filename[len - isalen], isastr, isalen) == 0)
384 		p -= isalen;
385 
386 	/*
387 	 * "/platform/mumblefrotz" + " " + MOD_DEFPATH
388 	 */
389 	len += (p - filename) + 1 + strlen(MOD_DEFPATH) + 1;
390 	(void) strncpy(path, filename, p - filename);
391 }
392