1 /*-
2 * Copyright (c) KATO Takenori, 1997, 1998.
3 *
4 * All rights reserved. Unpublished rights reserved under the copyright
5 * laws of Japan.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer as
13 * the first lines of this file unmodified.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 */
29
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD: releng/11.2/sys/amd64/amd64/initcpu.c 334152 2018-05-24 13:17:24Z kib $");
32
33 #include "opt_cpu.h"
34
35 #include <sys/param.h>
36 #include <sys/kernel.h>
37 #include <sys/pcpu.h>
38 #include <sys/systm.h>
39 #include <sys/sysctl.h>
40
41 #include <machine/cputypes.h>
42 #include <machine/md_var.h>
43 #include <machine/specialreg.h>
44
45 #include <vm/vm.h>
46 #include <vm/pmap.h>
47
48 static int hw_instruction_sse;
49 SYSCTL_INT(_hw, OID_AUTO, instruction_sse, CTLFLAG_RD,
50 &hw_instruction_sse, 0, "SIMD/MMX2 instructions available in CPU");
51 static int lower_sharedpage_init;
52 int hw_lower_amd64_sharedpage;
53 SYSCTL_INT(_hw, OID_AUTO, lower_amd64_sharedpage, CTLFLAG_RDTUN,
54 &hw_lower_amd64_sharedpage, 0,
55 "Lower sharedpage to work around Ryzen issue with executing code near the top of user memory");
56 /*
57 * -1: automatic (default)
58 * 0: keep enable CLFLUSH
59 * 1: force disable CLFLUSH
60 */
61 static int hw_clflush_disable = -1;
62
63 static void
64 init_amd(void)
65 {
66 uint64_t msr;
67
68 /*
69 * Work around Erratum 721 for Family 10h and 12h processors.
70 * These processors may incorrectly update the stack pointer
71 * after a long series of push and/or near-call instructions,
72 * or a long series of pop and/or near-return instructions.
73 *
74 * http://support.amd.com/us/Processor_TechDocs/41322_10h_Rev_Gd.pdf
75 * http://support.amd.com/us/Processor_TechDocs/44739_12h_Rev_Gd.pdf
76 *
77 * Hypervisors do not provide access to the errata MSR,
78 * causing #GP exception on attempt to apply the errata. The
79 * MSR write shall be done on host and persist globally
80 * anyway, so do not try to do it when under virtualization.
81 */
82 switch (CPUID_TO_FAMILY(cpu_id)) {
83 case 0x10:
84 case 0x12:
85 if ((cpu_feature2 & CPUID2_HV) == 0)
86 wrmsr(0xc0011029, rdmsr(0xc0011029) | 1);
87 break;
88 }
89
90 /*
91 * BIOS may fail to set InitApicIdCpuIdLo to 1 as it should per BKDG.
92 * So, do it here or otherwise some tools could be confused by
93 * Initial Local APIC ID reported with CPUID Function 1 in EBX.
94 */
95 if (CPUID_TO_FAMILY(cpu_id) == 0x10) {
96 if ((cpu_feature2 & CPUID2_HV) == 0) {
97 msr = rdmsr(MSR_NB_CFG1);
98 msr |= (uint64_t)1 << 54;
99 wrmsr(MSR_NB_CFG1, msr);
100 }
101 }
102
103 /*
104 * BIOS may configure Family 10h processors to convert WC+ cache type
105 * to CD. That can hurt performance of guest VMs using nested paging.
106 * The relevant MSR bit is not documented in the BKDG,
107 * the fix is borrowed from Linux.
108 */
109 if (CPUID_TO_FAMILY(cpu_id) == 0x10) {
110 if ((cpu_feature2 & CPUID2_HV) == 0) {
111 msr = rdmsr(0xc001102a);
112 msr &= ~((uint64_t)1 << 24);
113 wrmsr(0xc001102a, msr);
114 }
115 }
116
117 /*
118 * Work around Erratum 793: Specific Combination of Writes to Write
119 * Combined Memory Types and Locked Instructions May Cause Core Hang.
120 * See Revision Guide for AMD Family 16h Models 00h-0Fh Processors,
121 * revision 3.04 or later, publication 51810.
122 */
123 if (CPUID_TO_FAMILY(cpu_id) == 0x16 && CPUID_TO_MODEL(cpu_id) <= 0xf) {
124 if ((cpu_feature2 & CPUID2_HV) == 0) {
125 msr = rdmsr(0xc0011020);
126 msr |= (uint64_t)1 << 15;
127 wrmsr(0xc0011020, msr);
128 }
129 }
130
131 /*
132 * Work around a problem on Ryzen that is triggered by executing
133 * code near the top of user memory, in our case the signal
134 * trampoline code in the shared page on amd64.
135 *
136 * This function is executed once for the BSP before tunables take
137 * effect so the value determined here can be overridden by the
138 * tunable. This function is then executed again for each AP and
139 * also on resume. Set a flag the first time so that value set by
140 * the tunable is not overwritten.
141 *
142 * The stepping and/or microcode versions should be checked after
143 * this issue is fixed by AMD so that we don't use this mode if not
144 * needed.
145 */
146 if (lower_sharedpage_init == 0) {
147 lower_sharedpage_init = 1;
148 if (CPUID_TO_FAMILY(cpu_id) == 0x17) {
149 hw_lower_amd64_sharedpage = 1;
150 }
151 }
152 }
153
154 /*
155 * Initialize special VIA features
156 */
157 static void
158 init_via(void)
159 {
160 u_int regs[4], val;
161
162 /*
163 * Check extended CPUID for PadLock features.
164 *
165 * http://www.via.com.tw/en/downloads/whitepapers/initiatives/padlock/programming_guide.pdf
166 */
167 do_cpuid(0xc0000000, regs);
168 if (regs[0] >= 0xc0000001) {
169 do_cpuid(0xc0000001, regs);
170 val = regs[3];
171 } else
172 return;
173
174 /* Enable RNG if present. */
175 if ((val & VIA_CPUID_HAS_RNG) != 0) {
176 via_feature_rng = VIA_HAS_RNG;
177 wrmsr(0x110B, rdmsr(0x110B) | VIA_CPUID_DO_RNG);
178 }
179
180 /* Enable PadLock if present. */
181 if ((val & VIA_CPUID_HAS_ACE) != 0)
182 via_feature_xcrypt |= VIA_HAS_AES;
183 if ((val & VIA_CPUID_HAS_ACE2) != 0)
184 via_feature_xcrypt |= VIA_HAS_AESCTR;
185 if ((val & VIA_CPUID_HAS_PHE) != 0)
186 via_feature_xcrypt |= VIA_HAS_SHA;
187 if ((val & VIA_CPUID_HAS_PMM) != 0)
188 via_feature_xcrypt |= VIA_HAS_MM;
189 if (via_feature_xcrypt != 0)
190 wrmsr(0x1107, rdmsr(0x1107) | (1 << 28));
191 }
192
193 /*
194 * Initialize CPU control registers
195 */
196 void
197 initializecpu(void)
198 {
199 uint64_t msr;
200 uint32_t cr4;
201
202 cr4 = rcr4();
203 if ((cpu_feature & CPUID_XMM) && (cpu_feature & CPUID_FXSR)) {
204 cr4 |= CR4_FXSR | CR4_XMM;
205 cpu_fxsr = hw_instruction_sse = 1;
206 }
207 if (cpu_stdext_feature & CPUID_STDEXT_FSGSBASE)
208 cr4 |= CR4_FSGSBASE;
209
210 /*
211 * Postpone enabling the SMEP on the boot CPU until the page
212 * tables are switched from the boot loader identity mapping
213 * to the kernel tables. The boot loader enables the U bit in
214 * its tables.
215 */
216 if (!IS_BSP() && (cpu_stdext_feature & CPUID_STDEXT_SMEP))
217 cr4 |= CR4_SMEP;
218 load_cr4(cr4);
219 if ((amd_feature & AMDID_NX) != 0) {
220 msr = rdmsr(MSR_EFER) | EFER_NXE;
221 wrmsr(MSR_EFER, msr);
222 pg_nx = PG_NX;
223 }
224 hw_ibrs_recalculate();
225 hw_ssb_recalculate(false);
226 switch (cpu_vendor_id) {
227 case CPU_VENDOR_AMD:
228 init_amd();
229 break;
230 case CPU_VENDOR_CENTAUR:
231 init_via();
232 break;
233 }
234 }
235
236 void
237 initializecpucache(void)
238 {
239
240 /*
241 * CPUID with %eax = 1, %ebx returns
242 * Bits 15-8: CLFLUSH line size
243 * (Value * 8 = cache line size in bytes)
244 */
245 if ((cpu_feature & CPUID_CLFSH) != 0)
246 cpu_clflush_line_size = ((cpu_procinfo >> 8) & 0xff) * 8;
247 /*
248 * XXXKIB: (temporary) hack to work around traps generated
249 * when CLFLUSHing APIC register window under virtualization
250 * environments. These environments tend to disable the
251 * CPUID_SS feature even though the native CPU supports it.
252 */
253 TUNABLE_INT_FETCH("hw.clflush_disable", &hw_clflush_disable);
254 if (vm_guest != VM_GUEST_NO && hw_clflush_disable == -1) {
255 cpu_feature &= ~CPUID_CLFSH;
256 cpu_stdext_feature &= ~CPUID_STDEXT_CLFLUSHOPT;
257 }
258
259 /*
260 * The kernel's use of CLFLUSH{,OPT} can be disabled manually
261 * by setting the hw.clflush_disable tunable.
262 */
263 if (hw_clflush_disable == 1) {
264 cpu_feature &= ~CPUID_CLFSH;
265 cpu_stdext_feature &= ~CPUID_STDEXT_CLFLUSHOPT;
266 }
267 }
Cache object: 7d1c2e1377034f60b6875f2831494a9c
|