1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
|
/* SPDX-License-Identifier: GPL-2.0-only */
#include <acpi/acpigen.h>
#include <acpi/acpi.h>
#include <assert.h>
#include <console/console.h>
#include <console/debug.h>
#include <cpu/cpu.h>
#include <cpu/intel/common/common.h>
#include <cpu/intel/em64t101_save_state.h>
#include <cpu/intel/microcode.h>
#include <cpu/intel/smm_reloc.h>
#include <cpu/intel/turbo.h>
#include <cpu/x86/lapic.h>
#include <cpu/x86/mp.h>
#include <cpu/x86/mtrr.h>
#include <intelblocks/cpulib.h>
#include <intelblocks/mp_init.h>
#include <soc/cpu.h>
#include <soc/msr.h>
#include <soc/soc_util.h>
#include <soc/smmrelocate.h>
#include <soc/util.h>
#include "chip.h"
static const void *microcode_patch;
static const config_t *chip_config = NULL;
static void xeon_configure_mca(void)
{
msr_t msr;
struct cpuid_result cpuid_regs;
/*
* Check feature flag in CPUID.(EAX=1):EDX[7]==1 MCE
* and CPUID.(EAX=1):EDX[14]==1 MCA
*/
cpuid_regs = cpuid(1);
if ((cpuid_regs.edx & (1 << 7 | 1 << 14)) != (1 << 7 | 1 << 14))
return;
msr = rdmsr(IA32_MCG_CAP);
if (msr.lo & IA32_MCG_CAP_CTL_P_MASK) {
/* Enable all error logging */
msr.lo = msr.hi = 0xffffffff;
wrmsr(IA32_MCG_CTL, msr);
}
mca_configure();
}
/*
* On server platforms the FIT mechanism only updates the microcode on
* the BSP. Loading MCU on AP in parallel seems to fail in 10% of the cases
* so do it serialized.
*/
void get_microcode_info(const void **microcode, int *parallel)
{
*microcode = intel_microcode_find();
*parallel = 0;
}
static void each_cpu_init(struct device *cpu)
{
msr_t msr;
printk(BIOS_SPEW, "%s dev: %s, cpu: %d, apic_id: 0x%x\n",
__func__, dev_path(cpu), cpu_index(), cpu->path.apic.apic_id);
setup_lapic();
/*
* Set HWP base feature, EPP reg enumeration, lock thermal and msr
* This is package level MSR. Need to check if it updates correctly on
* multi-socket platform.
*/
msr = rdmsr(MSR_MISC_PWR_MGMT);
if (!(msr.lo & LOCK_MISC_PWR_MGMT_MSR)) { /* if already locked skip update */
msr.lo = (HWP_ENUM_ENABLE | HWP_EPP_ENUM_ENABLE | LOCK_MISC_PWR_MGMT_MSR |
LOCK_THERM_INT);
wrmsr(MSR_MISC_PWR_MGMT, msr);
}
/* Enable Fast Strings */
msr = rdmsr(IA32_MISC_ENABLE);
msr.lo |= FAST_STRINGS_ENABLE_BIT;
wrmsr(IA32_MISC_ENABLE, msr);
/* Enable Turbo */
enable_turbo();
/* Enable speed step. */
if (get_turbo_state() == TURBO_ENABLED) {
msr = rdmsr(IA32_MISC_ENABLE);
msr.lo |= SPEED_STEP_ENABLE_BIT;
wrmsr(IA32_MISC_ENABLE, msr);
}
/* Clear out pending MCEs */
xeon_configure_mca();
/* Enable Vmx */
set_vmx_and_lock();
}
static struct device_operations cpu_dev_ops = {
.init = each_cpu_init,
};
static const struct cpu_device_id cpu_table[] = {
{X86_VENDOR_INTEL, CPUID_COOPERLAKE_SP_A0},
{X86_VENDOR_INTEL, CPUID_COOPERLAKE_SP_A1},
{0, 0},
};
static const struct cpu_driver driver __cpu_driver = {
.ops = &cpu_dev_ops,
.id_table = cpu_table,
};
static void set_max_turbo_freq(void)
{
msr_t msr, perf_ctl;
FUNC_ENTER();
perf_ctl.hi = 0;
/* Check for configurable TDP option */
if (get_turbo_state() == TURBO_ENABLED) {
msr = rdmsr(MSR_TURBO_RATIO_LIMIT);
perf_ctl.lo = (msr.lo & 0xff) << 8;
} else if (cpu_config_tdp_levels()) {
/* Set to nominal TDP ratio */
msr = rdmsr(MSR_CONFIG_TDP_NOMINAL);
perf_ctl.lo = (msr.lo & 0xff) << 8;
} else {
/* Platform Info bits 15:8 give max ratio */
msr = rdmsr(MSR_PLATFORM_INFO);
perf_ctl.lo = msr.lo & 0xff00;
}
wrmsr(IA32_PERF_CTL, perf_ctl);
printk(BIOS_DEBUG, "cpu: frequency set to %d\n",
((perf_ctl.lo >> 8) & 0xff) * CONFIG_CPU_BCLK_MHZ);
FUNC_EXIT();
}
/*
* Do essential initialization tasks before APs can be fired up
*/
static void pre_mp_init(void)
{
x86_setup_mtrrs_with_detect();
x86_mtrr_check();
}
static int get_thread_count(void)
{
unsigned int num_phys = 0, num_virts = 0;
cpu_read_topology(&num_phys, &num_virts);
printk(BIOS_SPEW, "Detected %u cores and %u threads\n", num_phys, num_virts);
/*
* Currently we do not know a way to figure out how many CPUs we have total
* on multi-socketed. So we pretend all sockets are populated with CPUs with
* same thread/core fusing.
* TODO: properly figure out number of active sockets OR refactor MPinit code
* to remove requirements of having to know total number of CPUs in advance.
*/
return num_virts * CONFIG_MAX_SOCKET;
}
static void post_mp_init(void)
{
/* Set Max Ratio */
set_max_turbo_freq();
if (CONFIG(HAVE_SMI_HANDLER))
global_smi_enable();
}
static const struct mp_ops mp_ops = {
.pre_mp_init = pre_mp_init,
.get_cpu_count = get_thread_count,
.get_smm_info = get_smm_info,
.pre_mp_smm_init = smm_initialize,
.relocation_handler = smm_relocation_handler,
.get_microcode_info = get_microcode_info,
.post_mp_init = post_mp_init,
};
void cpx_init_cpus(struct device *dev)
{
microcode_patch = intel_microcode_find();
if (!microcode_patch)
printk(BIOS_ERR, "microcode not found in CBFS!\n");
intel_microcode_load_unlocked(microcode_patch);
if (mp_init_with_smm(dev->link_list, &mp_ops) < 0)
printk(BIOS_ERR, "MP initialization failure.\n");
/*
* chip_config is used in cpu device callback. Other than cpu 0,
* rest of the CPU devices do not have chip_info updated.
*/
chip_config = dev->chip_info;
/* update numa domain for all cpu devices */
xeonsp_init_cpu_config();
}
|