xref: /titanic_51/usr/src/uts/i86pc/os/mlsetup.c (revision 63251bc7f1ca38259078c48e316fee4ed66d4e93)
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/kobj.h>
57 #include <sys/kobj_lex.h>
58 #include <sys/pci_cfgspace.h>
59 #ifdef __xpv
60 #include <sys/hypervisor.h>
61 #endif
62 
63 /*
64  * some globals for patching the result of cpuid
65  * to solve problems w/ creative cpu vendors
66  */
67 
68 extern uint32_t cpuid_feature_ecx_include;
69 extern uint32_t cpuid_feature_ecx_exclude;
70 extern uint32_t cpuid_feature_edx_include;
71 extern uint32_t cpuid_feature_edx_exclude;
72 
73 /*
74  * Dummy spl priority masks
75  */
76 static unsigned char dummy_cpu_pri[MAXIPL + 1] = {
77 	0xf, 0xf, 0xf, 0xf, 0xf, 0xf, 0xf, 0xf,
78 	0xf, 0xf, 0xf, 0xf, 0xf, 0xf, 0xf, 0xf, 0xf
79 };
80 
81 
82 static uint32_t
83 bootprop_getval(char *name)
84 {
85 	char prop[32];
86 	u_longlong_t ll;
87 	extern struct bootops *bootops;
88 	if ((BOP_GETPROPLEN(bootops, name) > sizeof (prop)) ||
89 	    (BOP_GETPROP(bootops, name, prop) < 0) ||
90 	    (kobj_getvalue(prop, &ll) == -1))
91 		return (0);
92 	return ((uint32_t)ll);
93 }
94 
95 /*
96  * Setup routine called right before main(). Interposing this function
97  * before main() allows us to call it in a machine-independent fashion.
98  */
99 void
100 mlsetup(struct regs *rp)
101 {
102 	extern struct classfuncs sys_classfuncs;
103 	extern disp_t cpu0_disp;
104 	extern char t0stack[];
105 
106 	ASSERT_STACK_ALIGNED();
107 
108 	/*
109 	 * initialize cpu_self
110 	 */
111 	cpu[0]->cpu_self = cpu[0];
112 
113 #if defined(__xpv)
114 	/*
115 	 * Point at the hypervisor's virtual cpu structure
116 	 */
117 	cpu[0]->cpu_m.mcpu_vcpu_info = &HYPERVISOR_shared_info->vcpu_info[0];
118 #endif
119 
120 	/*
121 	 * Set up dummy cpu_pri_data values till psm spl code is
122 	 * installed.  This allows splx() to work on amd64.
123 	 */
124 
125 	cpu[0]->cpu_pri_data = dummy_cpu_pri;
126 
127 	/*
128 	 * check if we've got special bits to clear or set
129 	 * when checking cpu features
130 	 */
131 
132 	cpuid_feature_ecx_include =
133 	    bootprop_getval("cpuid_feature_ecx_include");
134 	cpuid_feature_ecx_exclude =
135 	    bootprop_getval("cpuid_feature_ecx_exclude");
136 	cpuid_feature_edx_include =
137 	    bootprop_getval("cpuid_feature_edx_include");
138 	cpuid_feature_edx_exclude =
139 	    bootprop_getval("cpuid_feature_edx_exclude");
140 
141 	/*
142 	 * The first lightweight pass (pass0) through the cpuid data
143 	 * was done in locore before mlsetup was called.  Do the next
144 	 * pass in C code.
145 	 *
146 	 * The x86_feature bits are set here on the basis of the capabilities
147 	 * of the boot CPU.  Note that if we choose to support CPUs that have
148 	 * different feature sets (at which point we would almost certainly
149 	 * want to set the feature bits to correspond to the feature
150 	 * minimum) this value may be altered.
151 	 */
152 	x86_feature = cpuid_pass1(cpu[0]);
153 
154 	/*
155 	 * Initialize idt0, gdt0, ldt0_default, ktss0 and dftss.
156 	 */
157 	init_desctbls();
158 
159 #if !defined(__xpv)
160 
161 	/*
162 	 * Patch the tsc_read routine with appropriate set of instructions,
163 	 * depending on the processor family and architecure, to read the
164 	 * time-stamp counter while ensuring no out-of-order execution.
165 	 * Patch it while the kernel text is still writable.
166 	 *
167 	 * Note: tsc_read is not patched for intel processors whose family
168 	 * is >6 and for amd whose family >f (in case they don't support rdtscp
169 	 * instruction, unlikely). By default tsc_read will use cpuid for
170 	 * serialization in such cases. The following code needs to be
171 	 * revisited if intel processors of family >= f retains the
172 	 * instruction serialization nature of mfence instruction.
173 	 * Note: tsc_read is not patched for x86 processors which do
174 	 * not support "mfence". By default tsc_read will use cpuid for
175 	 * serialization in such cases.
176 	 *
177 	 * The Xen hypervisor does not correctly report whether rdtscp is
178 	 * supported or not, so we must assume that it is not.
179 	 */
180 	if (get_hwenv() != HW_XEN_HVM && (x86_feature & X86_TSCP))
181 		patch_tsc_read(X86_HAVE_TSCP);
182 	else if (cpuid_getvendor(CPU) == X86_VENDOR_AMD &&
183 	    cpuid_getfamily(CPU) <= 0xf && (x86_feature & X86_SSE2) != 0)
184 		patch_tsc_read(X86_TSC_MFENCE);
185 	else if (cpuid_getvendor(CPU) == X86_VENDOR_Intel &&
186 	    cpuid_getfamily(CPU) <= 6 && (x86_feature & X86_SSE2) != 0)
187 		patch_tsc_read(X86_TSC_LFENCE);
188 
189 #endif	/* !__xpv */
190 
191 #if defined(__i386) && !defined(__xpv)
192 	/*
193 	 * Some i386 processors do not implement the rdtsc instruction,
194 	 * or at least they do not implement it correctly. Patch them to
195 	 * return 0.
196 	 */
197 	if ((x86_feature & X86_TSC) == 0)
198 		patch_tsc_read(X86_NO_TSC);
199 #endif	/* __i386 && !__xpv */
200 
201 #if defined(__amd64) && !defined(__xpv)
202 	patch_memops(cpuid_getvendor(CPU));
203 #endif	/* __amd64 && !__xpv */
204 
205 #if !defined(__xpv)
206 	/* XXPV	what, if anything, should be dorked with here under xen? */
207 
208 	/*
209 	 * While we're thinking about the TSC, let's set up %cr4 so that
210 	 * userland can issue rdtsc, and initialize the TSC_AUX value
211 	 * (the cpuid) for the rdtscp instruction on appropriately
212 	 * capable hardware.
213 	 */
214 	if (x86_feature & X86_TSC)
215 		setcr4(getcr4() & ~CR4_TSD);
216 
217 	if (x86_feature & X86_TSCP)
218 		(void) wrmsr(MSR_AMD_TSCAUX, 0);
219 
220 	if (x86_feature & X86_DE)
221 		setcr4(getcr4() | CR4_DE);
222 #endif /* __xpv */
223 
224 	/*
225 	 * initialize t0
226 	 */
227 	t0.t_stk = (caddr_t)rp - MINFRAME;
228 	t0.t_stkbase = t0stack;
229 	t0.t_pri = maxclsyspri - 3;
230 	t0.t_schedflag = TS_LOAD | TS_DONT_SWAP;
231 	t0.t_procp = &p0;
232 	t0.t_plockp = &p0lock.pl_lock;
233 	t0.t_lwp = &lwp0;
234 	t0.t_forw = &t0;
235 	t0.t_back = &t0;
236 	t0.t_next = &t0;
237 	t0.t_prev = &t0;
238 	t0.t_cpu = cpu[0];
239 	t0.t_disp_queue = &cpu0_disp;
240 	t0.t_bind_cpu = PBIND_NONE;
241 	t0.t_bind_pset = PS_NONE;
242 	t0.t_bindflag = (uchar_t)default_binding_mode;
243 	t0.t_cpupart = &cp_default;
244 	t0.t_clfuncs = &sys_classfuncs.thread;
245 	t0.t_copyops = NULL;
246 	THREAD_ONPROC(&t0, CPU);
247 
248 	lwp0.lwp_thread = &t0;
249 	lwp0.lwp_regs = (void *)rp;
250 	lwp0.lwp_procp = &p0;
251 	t0.t_tid = p0.p_lwpcnt = p0.p_lwprcnt = p0.p_lwpid = 1;
252 
253 	p0.p_exec = NULL;
254 	p0.p_stat = SRUN;
255 	p0.p_flag = SSYS;
256 	p0.p_tlist = &t0;
257 	p0.p_stksize = 2*PAGESIZE;
258 	p0.p_stkpageszc = 0;
259 	p0.p_as = &kas;
260 	p0.p_lockp = &p0lock;
261 	p0.p_brkpageszc = 0;
262 	p0.p_t1_lgrpid = LGRP_NONE;
263 	p0.p_tr_lgrpid = LGRP_NONE;
264 	sigorset(&p0.p_ignore, &ignoredefault);
265 
266 	CPU->cpu_thread = &t0;
267 	bzero(&cpu0_disp, sizeof (disp_t));
268 	CPU->cpu_disp = &cpu0_disp;
269 	CPU->cpu_disp->disp_cpu = CPU;
270 	CPU->cpu_dispthread = &t0;
271 	CPU->cpu_idle_thread = &t0;
272 	CPU->cpu_flags = CPU_READY | CPU_RUNNING | CPU_EXISTS | CPU_ENABLE;
273 	CPU->cpu_dispatch_pri = t0.t_pri;
274 
275 	CPU->cpu_id = 0;
276 
277 	CPU->cpu_pri = 12;		/* initial PIL for the boot CPU */
278 
279 	/*
280 	 * The kernel doesn't use LDTs unless a process explicitly requests one.
281 	 */
282 	p0.p_ldt_desc = null_sdesc;
283 
284 	/*
285 	 * Initialize thread/cpu microstate accounting
286 	 */
287 	init_mstate(&t0, LMS_SYSTEM);
288 	init_cpu_mstate(CPU, CMS_SYSTEM);
289 
290 	/*
291 	 * Initialize lists of available and active CPUs.
292 	 */
293 	cpu_list_init(CPU);
294 
295 	pg_cpu_bootstrap(CPU);
296 
297 	/*
298 	 * Now that we have taken over the GDT, IDT and have initialized
299 	 * active CPU list it's time to inform kmdb if present.
300 	 */
301 	if (boothowto & RB_DEBUG)
302 		kdi_idt_sync();
303 
304 	/*
305 	 * If requested (boot -d) drop into kmdb.
306 	 *
307 	 * This must be done after cpu_list_init() on the 64-bit kernel
308 	 * since taking a trap requires that we re-compute gsbase based
309 	 * on the cpu list.
310 	 */
311 	if (boothowto & RB_DEBUGENTER)
312 		kmdb_enter();
313 
314 	cpu_vm_data_init(CPU);
315 
316 	/* lgrp_init() needs PCI config space access */
317 #if defined(__xpv)
318 	if (DOMAIN_IS_INITDOMAIN(xen_info))
319 		pci_cfgspace_init();
320 #else
321 	pci_cfgspace_init();
322 #endif
323 
324 	rp->r_fp = 0;	/* terminate kernel stack traces! */
325 
326 	prom_init("kernel", (void *)NULL);
327 
328 	boot_ncpus = bootprop_getval("boot-ncpus");
329 
330 	if (boot_ncpus <= 0 || boot_ncpus > NCPU)
331 		boot_ncpus = NCPU;
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