2 * QEMU fw_cfg helpers (X86 specific)
4 * Copyright (c) 2019 Red Hat, Inc.
7 * Philippe Mathieu-Daudé <philmd@redhat.com>
9 * SPDX-License-Identifier: GPL-2.0-or-later
11 * This work is licensed under the terms of the GNU GPL, version 2 or later.
12 * See the COPYING file in the top-level directory.
15 #include "qemu/osdep.h"
16 #include "sysemu/numa.h"
17 #include "hw/acpi/acpi.h"
18 #include "hw/acpi/aml-build.h"
19 #include "hw/firmware/smbios.h"
20 #include "hw/i386/fw_cfg.h"
21 #include "hw/timer/hpet.h"
22 #include "hw/nvram/fw_cfg.h"
23 #include "e820_memory_layout.h"
24 #include "kvm/kvm_i386.h"
25 #include "qapi/error.h"
26 #include CONFIG_DEVICES
28 struct hpet_fw_config hpet_cfg
= {.count
= UINT8_MAX
};
30 const char *fw_cfg_arch_key_name(uint16_t key
)
35 } fw_cfg_arch_wellknown_keys
[] = {
36 {FW_CFG_ACPI_TABLES
, "acpi_tables"},
37 {FW_CFG_SMBIOS_ENTRIES
, "smbios_entries"},
38 {FW_CFG_IRQ0_OVERRIDE
, "irq0_override"},
39 {FW_CFG_E820_TABLE
, "e820_table"},
40 {FW_CFG_HPET
, "hpet"},
43 for (size_t i
= 0; i
< ARRAY_SIZE(fw_cfg_arch_wellknown_keys
); i
++) {
44 if (fw_cfg_arch_wellknown_keys
[i
].key
== key
) {
45 return fw_cfg_arch_wellknown_keys
[i
].name
;
51 void fw_cfg_build_smbios(MachineState
*ms
, FWCfgState
*fw_cfg
)
54 uint8_t *smbios_tables
, *smbios_anchor
;
55 size_t smbios_tables_len
, smbios_anchor_len
;
56 struct smbios_phys_mem_area
*mem_array
;
57 unsigned i
, array_count
;
58 X86CPU
*cpu
= X86_CPU(ms
->possible_cpus
->cpus
[0].cpu
);
60 /* tell smbios about cpuid version and features */
61 smbios_set_cpuid(cpu
->env
.cpuid_version
, cpu
->env
.features
[FEAT_1_EDX
]);
63 smbios_tables
= smbios_get_table_legacy(ms
, &smbios_tables_len
);
65 fw_cfg_add_bytes(fw_cfg
, FW_CFG_SMBIOS_ENTRIES
,
66 smbios_tables
, smbios_tables_len
);
69 /* build the array of physical mem area from e820 table */
70 mem_array
= g_malloc0(sizeof(*mem_array
) * e820_get_num_entries());
71 for (i
= 0, array_count
= 0; i
< e820_get_num_entries(); i
++) {
74 if (e820_get_entry(i
, E820_RAM
, &addr
, &len
)) {
75 mem_array
[array_count
].address
= addr
;
76 mem_array
[array_count
].length
= len
;
80 smbios_get_tables(ms
, mem_array
, array_count
,
81 &smbios_tables
, &smbios_tables_len
,
82 &smbios_anchor
, &smbios_anchor_len
,
87 fw_cfg_add_file(fw_cfg
, "etc/smbios/smbios-tables",
88 smbios_tables
, smbios_tables_len
);
89 fw_cfg_add_file(fw_cfg
, "etc/smbios/smbios-anchor",
90 smbios_anchor
, smbios_anchor_len
);
95 FWCfgState
*fw_cfg_arch_create(MachineState
*ms
,
97 uint16_t apic_id_limit
)
100 uint64_t *numa_fw_cfg
;
102 MachineClass
*mc
= MACHINE_GET_CLASS(ms
);
103 const CPUArchIdList
*cpus
= mc
->possible_cpu_arch_ids(ms
);
104 int nb_numa_nodes
= ms
->numa_state
->num_nodes
;
106 fw_cfg
= fw_cfg_init_io_dma(FW_CFG_IO_BASE
, FW_CFG_IO_BASE
+ 4,
107 &address_space_memory
);
108 fw_cfg_add_i16(fw_cfg
, FW_CFG_NB_CPUS
, boot_cpus
);
110 /* FW_CFG_MAX_CPUS is a bit confusing/problematic on x86:
112 * For machine types prior to 1.8, SeaBIOS needs FW_CFG_MAX_CPUS for
113 * building MPTable, ACPI MADT, ACPI CPU hotplug and ACPI SRAT table,
114 * that tables are based on xAPIC ID and QEMU<->SeaBIOS interface
115 * for CPU hotplug also uses APIC ID and not "CPU index".
116 * This means that FW_CFG_MAX_CPUS is not the "maximum number of CPUs",
117 * but the "limit to the APIC ID values SeaBIOS may see".
119 * So for compatibility reasons with old BIOSes we are stuck with
120 * "etc/max-cpus" actually being apic_id_limit
122 fw_cfg_add_i16(fw_cfg
, FW_CFG_MAX_CPUS
, apic_id_limit
);
123 fw_cfg_add_i64(fw_cfg
, FW_CFG_RAM_SIZE
, ms
->ram_size
);
125 fw_cfg_add_bytes(fw_cfg
, FW_CFG_ACPI_TABLES
,
126 acpi_tables
, acpi_tables_len
);
128 fw_cfg_add_i32(fw_cfg
, FW_CFG_IRQ0_OVERRIDE
, 1);
130 fw_cfg_add_bytes(fw_cfg
, FW_CFG_E820_TABLE
,
131 &e820_reserve
, sizeof(e820_reserve
));
132 fw_cfg_add_file(fw_cfg
, "etc/e820", e820_table
,
133 sizeof(struct e820_entry
) * e820_get_num_entries());
135 fw_cfg_add_bytes(fw_cfg
, FW_CFG_HPET
, &hpet_cfg
, sizeof(hpet_cfg
));
136 /* allocate memory for the NUMA channel: one (64bit) word for the number
137 * of nodes, one word for each VCPU->node and one word for each node to
138 * hold the amount of memory.
140 numa_fw_cfg
= g_new0(uint64_t, 1 + apic_id_limit
+ nb_numa_nodes
);
141 numa_fw_cfg
[0] = cpu_to_le64(nb_numa_nodes
);
142 for (i
= 0; i
< cpus
->len
; i
++) {
143 unsigned int apic_id
= cpus
->cpus
[i
].arch_id
;
144 assert(apic_id
< apic_id_limit
);
145 numa_fw_cfg
[apic_id
+ 1] = cpu_to_le64(cpus
->cpus
[i
].props
.node_id
);
147 for (i
= 0; i
< nb_numa_nodes
; i
++) {
148 numa_fw_cfg
[apic_id_limit
+ 1 + i
] =
149 cpu_to_le64(ms
->numa_state
->nodes
[i
].node_mem
);
151 fw_cfg_add_bytes(fw_cfg
, FW_CFG_NUMA
, numa_fw_cfg
,
152 (1 + apic_id_limit
+ nb_numa_nodes
) *
153 sizeof(*numa_fw_cfg
));
158 void fw_cfg_build_feature_control(MachineState
*ms
, FWCfgState
*fw_cfg
)
160 X86CPU
*cpu
= X86_CPU(ms
->possible_cpus
->cpus
[0].cpu
);
161 CPUX86State
*env
= &cpu
->env
;
162 uint32_t unused
, ebx
, ecx
, edx
;
163 uint64_t feature_control_bits
= 0;
166 cpu_x86_cpuid(env
, 1, 0, &unused
, &unused
, &ecx
, &edx
);
167 if (ecx
& CPUID_EXT_VMX
) {
168 feature_control_bits
|= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX
;
171 if ((edx
& (CPUID_EXT2_MCE
| CPUID_EXT2_MCA
)) ==
172 (CPUID_EXT2_MCE
| CPUID_EXT2_MCA
) &&
173 (env
->mcg_cap
& MCG_LMCE_P
)) {
174 feature_control_bits
|= FEATURE_CONTROL_LMCE
;
177 if (env
->cpuid_level
>= 7) {
178 cpu_x86_cpuid(env
, 0x7, 0, &unused
, &ebx
, &ecx
, &unused
);
179 if (ebx
& CPUID_7_0_EBX_SGX
) {
180 feature_control_bits
|= FEATURE_CONTROL_SGX
;
182 if (ecx
& CPUID_7_0_ECX_SGX_LC
) {
183 feature_control_bits
|= FEATURE_CONTROL_SGX_LC
;
187 if (!feature_control_bits
) {
191 val
= g_malloc(sizeof(*val
));
192 *val
= cpu_to_le64(feature_control_bits
| FEATURE_CONTROL_LOCKED
);
193 fw_cfg_add_file(fw_cfg
, "etc/msr_feature_control", val
, sizeof(*val
));
196 void fw_cfg_add_acpi_dsdt(Aml
*scope
, FWCfgState
*fw_cfg
)
199 * when using port i/o, the 8-bit data register *always* overlaps
200 * with half of the 16-bit control register. Hence, the total size
201 * of the i/o region used is FW_CFG_CTL_SIZE; when using DMA, the
202 * DMA control register is located at FW_CFG_DMA_IO_BASE + 4
204 Object
*obj
= OBJECT(fw_cfg
);
205 uint8_t io_size
= object_property_get_bool(obj
, "dma_enabled", NULL
) ?
206 ROUND_UP(FW_CFG_CTL_SIZE
, 4) + sizeof(dma_addr_t
) :
208 Aml
*dev
= aml_device("FWCF");
209 Aml
*crs
= aml_resource_template();
211 aml_append(dev
, aml_name_decl("_HID", aml_string("QEMU0002")));
213 /* device present, functioning, decoding, not shown in UI */
214 aml_append(dev
, aml_name_decl("_STA", aml_int(0xB)));
217 aml_io(AML_DECODE16
, FW_CFG_IO_BASE
, FW_CFG_IO_BASE
, 0x01, io_size
));
219 aml_append(dev
, aml_name_decl("_CRS", crs
));
220 aml_append(scope
, dev
);