xref: /freebsd/sys/amd64/amd64/initcpu.c (revision 9119bafbaf52e5d86e5879e54b38c5aba3efa2e9)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) KATO Takenori, 1997, 1998.
5  *
6  * All rights reserved.  Unpublished rights reserved under the copyright
7  * laws of Japan.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer as
15  *    the first lines of this file unmodified.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_cpu.h"
36 
37 #include <sys/param.h>
38 #include <sys/kernel.h>
39 #include <sys/pcpu.h>
40 #include <sys/systm.h>
41 #include <sys/sysctl.h>
42 
43 #include <machine/cputypes.h>
44 #include <machine/md_var.h>
45 #include <machine/specialreg.h>
46 
47 #include <vm/vm.h>
48 #include <vm/pmap.h>
49 
50 static int	hw_instruction_sse;
51 SYSCTL_INT(_hw, OID_AUTO, instruction_sse, CTLFLAG_RD,
52     &hw_instruction_sse, 0, "SIMD/MMX2 instructions available in CPU");
53 static int	lower_sharedpage_init;
54 int		hw_lower_amd64_sharedpage;
55 SYSCTL_INT(_hw, OID_AUTO, lower_amd64_sharedpage, CTLFLAG_RDTUN,
56     &hw_lower_amd64_sharedpage, 0,
57    "Lower sharedpage to work around Ryzen issue with executing code near the top of user memory");
58 /*
59  * -1: automatic (default)
60  *  0: keep enable CLFLUSH
61  *  1: force disable CLFLUSH
62  */
63 static int	hw_clflush_disable = -1;
64 
65 static void
66 init_amd(void)
67 {
68 	uint64_t msr;
69 
70 	/*
71 	 * Work around Erratum 721 for Family 10h and 12h processors.
72 	 * These processors may incorrectly update the stack pointer
73 	 * after a long series of push and/or near-call instructions,
74 	 * or a long series of pop and/or near-return instructions.
75 	 *
76 	 * http://support.amd.com/us/Processor_TechDocs/41322_10h_Rev_Gd.pdf
77 	 * http://support.amd.com/us/Processor_TechDocs/44739_12h_Rev_Gd.pdf
78 	 *
79 	 * Hypervisors do not provide access to the errata MSR,
80 	 * causing #GP exception on attempt to apply the errata.  The
81 	 * MSR write shall be done on host and persist globally
82 	 * anyway, so do not try to do it when under virtualization.
83 	 */
84 	switch (CPUID_TO_FAMILY(cpu_id)) {
85 	case 0x10:
86 	case 0x12:
87 		if ((cpu_feature2 & CPUID2_HV) == 0)
88 			wrmsr(0xc0011029, rdmsr(0xc0011029) | 1);
89 		break;
90 	}
91 
92 	/*
93 	 * BIOS may fail to set InitApicIdCpuIdLo to 1 as it should per BKDG.
94 	 * So, do it here or otherwise some tools could be confused by
95 	 * Initial Local APIC ID reported with CPUID Function 1 in EBX.
96 	 */
97 	if (CPUID_TO_FAMILY(cpu_id) == 0x10) {
98 		if ((cpu_feature2 & CPUID2_HV) == 0) {
99 			msr = rdmsr(MSR_NB_CFG1);
100 			msr |= (uint64_t)1 << 54;
101 			wrmsr(MSR_NB_CFG1, msr);
102 		}
103 	}
104 
105 	/*
106 	 * BIOS may configure Family 10h processors to convert WC+ cache type
107 	 * to CD.  That can hurt performance of guest VMs using nested paging.
108 	 * The relevant MSR bit is not documented in the BKDG,
109 	 * the fix is borrowed from Linux.
110 	 */
111 	if (CPUID_TO_FAMILY(cpu_id) == 0x10) {
112 		if ((cpu_feature2 & CPUID2_HV) == 0) {
113 			msr = rdmsr(0xc001102a);
114 			msr &= ~((uint64_t)1 << 24);
115 			wrmsr(0xc001102a, msr);
116 		}
117 	}
118 
119 	/*
120 	 * Work around Erratum 793: Specific Combination of Writes to Write
121 	 * Combined Memory Types and Locked Instructions May Cause Core Hang.
122 	 * See Revision Guide for AMD Family 16h Models 00h-0Fh Processors,
123 	 * revision 3.04 or later, publication 51810.
124 	 */
125 	if (CPUID_TO_FAMILY(cpu_id) == 0x16 && CPUID_TO_MODEL(cpu_id) <= 0xf) {
126 		if ((cpu_feature2 & CPUID2_HV) == 0) {
127 			msr = rdmsr(MSR_LS_CFG);
128 			msr |= (uint64_t)1 << 15;
129 			wrmsr(MSR_LS_CFG, msr);
130 		}
131 	}
132 
133 	/* Ryzen erratas. */
134 	if (CPUID_TO_FAMILY(cpu_id) == 0x17 && CPUID_TO_MODEL(cpu_id) == 0x1 &&
135 	    (cpu_feature2 & CPUID2_HV) == 0) {
136 		/* 1021 */
137 		msr = rdmsr(0xc0011029);
138 		msr |= 0x2000;
139 		wrmsr(0xc0011029, msr);
140 
141 		/* 1033 */
142 		msr = rdmsr(MSR_LS_CFG);
143 		msr |= 0x10;
144 		wrmsr(MSR_LS_CFG, msr);
145 
146 		/* 1049 */
147 		msr = rdmsr(0xc0011028);
148 		msr |= 0x10;
149 		wrmsr(0xc0011028, msr);
150 
151 		/* 1095 */
152 		msr = rdmsr(MSR_LS_CFG);
153 		msr |= 0x200000000000000;
154 		wrmsr(MSR_LS_CFG, msr);
155 	}
156 
157 	/*
158 	 * Work around a problem on Ryzen that is triggered by executing
159 	 * code near the top of user memory, in our case the signal
160 	 * trampoline code in the shared page on amd64.
161 	 *
162 	 * This function is executed once for the BSP before tunables take
163 	 * effect so the value determined here can be overridden by the
164 	 * tunable.  This function is then executed again for each AP and
165 	 * also on resume.  Set a flag the first time so that value set by
166 	 * the tunable is not overwritten.
167 	 *
168 	 * The stepping and/or microcode versions should be checked after
169 	 * this issue is fixed by AMD so that we don't use this mode if not
170 	 * needed.
171 	 */
172 	if (lower_sharedpage_init == 0) {
173 		lower_sharedpage_init = 1;
174 		if (CPUID_TO_FAMILY(cpu_id) == 0x17 ||
175 		    CPUID_TO_FAMILY(cpu_id) == 0x18) {
176 			hw_lower_amd64_sharedpage = 1;
177 		}
178 	}
179 }
180 
181 /*
182  * Initialize special VIA features
183  */
184 static void
185 init_via(void)
186 {
187 	u_int regs[4], val;
188 
189 	/*
190 	 * Check extended CPUID for PadLock features.
191 	 *
192 	 * http://www.via.com.tw/en/downloads/whitepapers/initiatives/padlock/programming_guide.pdf
193 	 */
194 	do_cpuid(0xc0000000, regs);
195 	if (regs[0] >= 0xc0000001) {
196 		do_cpuid(0xc0000001, regs);
197 		val = regs[3];
198 	} else
199 		return;
200 
201 	/* Enable RNG if present. */
202 	if ((val & VIA_CPUID_HAS_RNG) != 0) {
203 		via_feature_rng = VIA_HAS_RNG;
204 		wrmsr(0x110B, rdmsr(0x110B) | VIA_CPUID_DO_RNG);
205 	}
206 
207 	/* Enable PadLock if present. */
208 	if ((val & VIA_CPUID_HAS_ACE) != 0)
209 		via_feature_xcrypt |= VIA_HAS_AES;
210 	if ((val & VIA_CPUID_HAS_ACE2) != 0)
211 		via_feature_xcrypt |= VIA_HAS_AESCTR;
212 	if ((val & VIA_CPUID_HAS_PHE) != 0)
213 		via_feature_xcrypt |= VIA_HAS_SHA;
214 	if ((val & VIA_CPUID_HAS_PMM) != 0)
215 		via_feature_xcrypt |= VIA_HAS_MM;
216 	if (via_feature_xcrypt != 0)
217 		wrmsr(0x1107, rdmsr(0x1107) | (1 << 28));
218 }
219 
220 /*
221  * Initialize CPU control registers
222  */
223 void
224 initializecpu(void)
225 {
226 	uint64_t msr;
227 	uint32_t cr4;
228 
229 	cr4 = rcr4();
230 	if ((cpu_feature & CPUID_XMM) && (cpu_feature & CPUID_FXSR)) {
231 		cr4 |= CR4_FXSR | CR4_XMM;
232 		cpu_fxsr = hw_instruction_sse = 1;
233 	}
234 	if (cpu_stdext_feature & CPUID_STDEXT_FSGSBASE)
235 		cr4 |= CR4_FSGSBASE;
236 
237 	if (cpu_stdext_feature2 & CPUID_STDEXT2_PKU)
238 		cr4 |= CR4_PKE;
239 
240 	/*
241 	 * If SMEP is present, we only need to flush RSB (by default)
242 	 * on context switches, to prevent cross-process ret2spec
243 	 * attacks.  Do it automatically if ibrs_disable is set, to
244 	 * complete the mitigation.
245 	 *
246 	 * Postpone enabling the SMEP on the boot CPU until the page
247 	 * tables are switched from the boot loader identity mapping
248 	 * to the kernel tables.  The boot loader enables the U bit in
249 	 * its tables.
250 	 */
251 	if (IS_BSP()) {
252 		if (cpu_stdext_feature & CPUID_STDEXT_SMEP &&
253 		    !TUNABLE_INT_FETCH(
254 		    "machdep.mitigations.cpu_flush_rsb_ctxsw",
255 		    &cpu_flush_rsb_ctxsw) &&
256 		    hw_ibrs_disable)
257 			cpu_flush_rsb_ctxsw = 1;
258 	} else {
259 		if (cpu_stdext_feature & CPUID_STDEXT_SMEP)
260 			cr4 |= CR4_SMEP;
261 		if (cpu_stdext_feature & CPUID_STDEXT_SMAP)
262 			cr4 |= CR4_SMAP;
263 	}
264 	load_cr4(cr4);
265 	if (IS_BSP() && (amd_feature & AMDID_NX) != 0) {
266 		msr = rdmsr(MSR_EFER) | EFER_NXE;
267 		wrmsr(MSR_EFER, msr);
268 		pg_nx = PG_NX;
269 	}
270 	hw_ibrs_recalculate(false);
271 	hw_ssb_recalculate(false);
272 	amd64_syscall_ret_flush_l1d_recalc();
273 	x86_rngds_mitg_recalculate(false);
274 	switch (cpu_vendor_id) {
275 	case CPU_VENDOR_AMD:
276 	case CPU_VENDOR_HYGON:
277 		init_amd();
278 		break;
279 	case CPU_VENDOR_CENTAUR:
280 		init_via();
281 		break;
282 	}
283 
284 	if ((amd_feature & AMDID_RDTSCP) != 0 ||
285 	    (cpu_stdext_feature2 & CPUID_STDEXT2_RDPID) != 0)
286 		wrmsr(MSR_TSC_AUX, PCPU_GET(cpuid));
287 }
288 
289 void
290 initializecpucache(void)
291 {
292 
293 	/*
294 	 * CPUID with %eax = 1, %ebx returns
295 	 * Bits 15-8: CLFLUSH line size
296 	 * 	(Value * 8 = cache line size in bytes)
297 	 */
298 	if ((cpu_feature & CPUID_CLFSH) != 0)
299 		cpu_clflush_line_size = ((cpu_procinfo >> 8) & 0xff) * 8;
300 	/*
301 	 * XXXKIB: (temporary) hack to work around traps generated
302 	 * when CLFLUSHing APIC register window under virtualization
303 	 * environments.  These environments tend to disable the
304 	 * CPUID_SS feature even though the native CPU supports it.
305 	 */
306 	TUNABLE_INT_FETCH("hw.clflush_disable", &hw_clflush_disable);
307 	if (vm_guest != VM_GUEST_NO && hw_clflush_disable == -1) {
308 		cpu_feature &= ~CPUID_CLFSH;
309 		cpu_stdext_feature &= ~CPUID_STDEXT_CLFLUSHOPT;
310 	}
311 
312 	/*
313 	 * The kernel's use of CLFLUSH{,OPT} can be disabled manually
314 	 * by setting the hw.clflush_disable tunable.
315 	 */
316 	if (hw_clflush_disable == 1) {
317 		cpu_feature &= ~CPUID_CLFSH;
318 		cpu_stdext_feature &= ~CPUID_STDEXT_CLFLUSHOPT;
319 	}
320 }
321