Support QEMU cpu models in MSHV accelerator
cover Support QEMU cpu models in MSHV accelerator
Hey all,
In the current MSHV accelerator code passing CPU features via the -cpu flag doesn't work as intended yet. When using the MSHV hypervisor we either silently discard the specified model/features and leave it up to the hypervisor to provide a sensible set of features or if the user selects -cpu host, the hypervisor might refuse to create a partition.
This changeset introduces a more comprehensive support for passing desired guest cpu features to the hypervisor. It's also a prerequisite for Live Migration support, in which we have to roundtrip CPU State explicitly
Known issues:
We will probably have to iterate a bit more on this, since recently support for CET_U/CET_S xstate bits has been introduced in QEMU, which doesn't harmonize with our current approach of configuring the hypervisor with static responses to cpuid queries.
Drive-by fixes:
- tiny fix in MAINTAINERS
- adding packed attribute to inlined UAPI structs
best,
magnus
Magnus Kulke (10):
MAINTAINERS: fix magnuskulke email-address include/hw/hyperv: add QEMU_PACKED to uapi structs accel/mshv: use mshv_create_partition_v2 payload target/i386/mshv: fix cpuid propagation bug target/i386/mshv: fix various cpuid traversal bugs target/i386/mshv: change cpuid mask to UINT32_MAX target/i386/mshv: set cpu model name on -cpu host target/i386: query mshv accel for supported cpuids target/i386/mshv: populate xsave area offsets target/i386/mshv: filter out CET bits in cpuid
MAINTAINERS | 4 +- accel/mshv/mshv-all.c | 35 ++++-- include/hw/hyperv/hvgdk_mini.h | 39 ++++-- include/hw/hyperv/hvhdk.h | 199 ++++++++++++++++++++++++++++- include/system/mshv.h | 3 + target/i386/cpu.c | 8 ++ target/i386/mshv/mshv-cpu.c | 221 ++++++++++++++++++++++++++------- 7 files changed, 438 insertions(+), 71 deletions(-)
Signature · 1 line
2.34.1
PATCH 1/10 MAINTAINERS: fix magnuskulke email-address
Consolidating email aliases.
--- MAINTAINERS | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
MAINTAINERS +2 -2
@@ -569,7 +569,7 @@ F: accel/stubs/whpx-stub.c F: include/system/whpx.h MSHV -M: Magnus Kulke <[email protected]> +M: Magnus Kulke <[email protected]> R: Wei Liu <[email protected]> S: Supported F: accel/mshv/
@@ -578,7 +578,7 @@ F: include/hw/hyperv/hvgdk*.h F: include/hw/hyperv/hvhdk*.h X86 MSHV CPUs -M: Magnus Kulke <[email protected]> +M: Magnus Kulke <[email protected]> R: Wei Liu <[email protected]> S: Supported F: target/i386/mshv/
Signature · 1 line
2.34.1
Re: [PATCH 01/10] MAINTAINERS: fix magnuskulke email-address
Quoted 22 lines · Magnus Kulke
Consolidating email aliases.
Signed-off-by Magnus Kulke <[email protected]>--- MAINTAINERS | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)MAINTAINERS +1
-1@@ -569,7 +569,7 @@ F: accel/stubs/whpx-stub.c F: include/system/whpx.h MSHV -M: Magnus Kulke <[email protected]> +M: Magnus Kulke <[email protected]> R: Wei Liu <[email protected]> S: Supported F: accel/mshv/
MAINTAINERS F: include/hw/hyperv/hvgdk*.h · line 584
Quoted 8 lines
MAINTAINERS +1
-1@@ -578,7 +578,7 @@ F: include/hw/hyperv/hvgdk*.h F: include/hw/hyperv/hvhdk*.h X86 MSHV CPUs -M: Magnus Kulke <[email protected]> +M: Magnus Kulke <[email protected]> R: Wei Liu <[email protected]> S: Supported F: target/i386/mshv/
Hi; I'm going to pull this MAINTAINERS patch out of this series and apply it via target-arm.next, since the incorrect email we currently have listed in the file bounces ("550 5.1.1 :
Recipient address rejected: User unknown in local recipient table").
thanks
-- PMM
Re: [PATCH 01/10] MAINTAINERS: fix magnuskulke email-address
Quoted 7 lines · Peter Maydell
Hi; I'm going to pull this
MAINTAINERSpatch out of this series and apply it via target-arm.next, since the incorrect email we currently have listed in the file bounces ("550 5.1.1 :
Recipient address rejected: User unknown in local recipient table").thanks
-- PMM
ack, thanks!
PATCH 2/10 include/hw/hyperv: add QEMU_PACKED to uapi structs
In reply to Magnus KulkeThe uapi definitions are marked with __packed hints in the kernel headers, since we want to keep the contract of the Microsoft Hypervisor ABI explicit, we should also added them in our vendored files, with a few notable exceptions where the attribute is a noop.
--- include/hw/hyperv/hvgdk_mini.h | 37 ++++++++++++++++++++++------------ include/hw/hyperv/hvhdk.h | 14 +++++++------ 2 files changed, 32 insertions(+), 19 deletions(-)
include/hw/hyperv/hvgdk_mini.h +24 -13
@@ -7,6 +7,8 @@ #ifndef HW_HYPERV_HVGDK_MINI_H #define HW_HYPERV_HVGDK_MINI_H +#include "qemu/compiler.h" + #define MSHV_IOCTL 0xB8 typedef enum hv_register_name {
@@ -211,7 +213,7 @@ enum hv_intercept_type {struct hv_u128 {uint64_t low_part;uint64_t high_part; -}; +}QEMU_PACKED ;union hv_x64_xmm_control_status_register {struct hv_u128 as_uint128;
@@ -326,13 +328,13 @@ typedef struct hv_x64_segment_register { };uint16_t attributes; }; -} hv_x64_segment_register; +}QEMU_PACKED hv_x64_segment_register;typedef struct hv_x64_table_register {uint16_t pad[3 ];uint16_t limit;uint64_t base; -}hv_x64_table_register ; +} QEMU_PACKED hv_x64_table_register;union hv_x64_fp_control_status_register {struct hv_u128 as_uint128;
@@ -416,6 +418,10 @@ typedef union hv_register_value {union hv_x64_register_sev_control sev_control; } hv_register_value; +/* +* This struct is __packed in the kernel. Since all members are naturally +* aligned, we can omit QEMU_PACKED to avoid address-of-packed-member warnings. +*/ typedef struct hv_register_assoc { uint32_t name; /* enum hv_register_name */ uint32_t reserved1;
@@ -439,7 +445,7 @@ typedef struct hv_input_get_vp_registers {uint8_t rsvd_z8;uint16_t rsvd_z16;uint32_t names[]; -} hv_input_get_vp_registers; +}QEMU_PACKED hv_input_get_vp_registers; typedef struct hv_input_set_vp_registers { uint64_t partition_id;
@@ -448,7 +454,7 @@ typedef struct hv_input_set_vp_registers {uint8_t rsvd_z8;uint16_t rsvd_z16;struct hv_register_assoc elements[]; -} hv_input_set_vp_registers; +}QEMU_PACKED hv_input_set_vp_registers;union hv_interrupt_control {uint64_t as_uint64;
@@ -468,7 +474,7 @@ struct hv_input_assert_virtual_interrupt {uint8_t target_vtl;uint8_t rsvd_z0;uint16_t rsvd_z1; -}; +}QEMU_PACKED ;/* /dev/mshv */ #define MSHV_CREATE_PARTITION _IOW(MSHV_IOCTL, 0x00, struct mshv_create_partition)
@@ -487,6 +493,11 @@ struct hv_input_assert_virtual_interrupt {******************************** */ +/* +* This struct is __packed in the kernel, but since all members are naturally +* aligned, so we can omit QEMU_PACKED to avoid address-of-packed-member +* warnings. +*/ struct hv_local_interrupt_controller_state {/* HV_X64_INTERRUPT_CONTROLLER_STATE */ uint32_t apic_id;
@@ -644,7 +655,7 @@ struct hv_x64_intercept_message_header {struct hv_x64_segment_register cs_segment;uint64_t rip;uint64_t rflags; -}; +}QEMU_PACKED ;union hv_x64_io_port_access_info {uint8_t as_uint8;
@@ -669,7 +680,7 @@ typedef struct hv_x64_io_port_intercept_message {uint64_t rcx;uint64_t rsi;uint64_t rdi; -} hv_x64_io_port_intercept_message; +}QEMU_PACKED hv_x64_io_port_intercept_message;union hv_x64_memory_access_info {uint8_t as_uint8;
@@ -692,7 +703,7 @@ struct hv_x64_memory_intercept_message {uint64_t guest_virtual_address;uint64_t guest_physical_address;uint8_t instruction_bytes[16 ]; -}; +}QEMU_PACKED ;union hv_message_flags {uint8_t asu8;
@@ -711,14 +722,14 @@ struct hv_message_header {uint64_t sender;union hv_port_id port; }; -}; +}QEMU_PACKED ;struct hv_message {struct hv_message_header header;union {uint64_t payload[HV_MESSAGE_PAYLOAD_QWORD_COUNT ]; } u; -}; +}QEMU_PACKED ;/* From github.com/rust-vmm/mshv-bindings/src/x86_64/regs.rs */
@@ -731,13 +742,13 @@ struct hv_cpuid_entry {uint32_t ecx;uint32_t edx;uint32_t padding[3 ]; -}; +}QEMU_PACKED ;struct hv_cpuid {uint32_t nent;uint32_t padding;struct hv_cpuid_entry entries[0 ]; -}; +}QEMU_PACKED ;#define IA32_MSR_TSC 0x00000010#define IA32_MSR_EFER 0xC0000080
include/hw/hyperv/hvhdk.h +8 -6
@@ -9,6 +9,8 @@#ifndef HW_HYPERV_HVHDK_H #define HW_HYPERV_HVHDK_H +#include "qemu/compiler.h" +#define HV_PARTITION_SYNTHETIC_PROCESSOR_FEATURES_BANKS 1 struct hv_input_set_partition_property {
@@ -16,7 +18,7 @@ struct hv_input_set_partition_property {uint32_t property_code;/* enum hv_partition_property_code */ uint32_t padding;uint64_t property_value; -}; +}QEMU_PACKED ;union hv_partition_synthetic_processor_features {uint64_t as_uint64[HV_PARTITION_SYNTHETIC_PROCESSOR_FEATURES_BANKS ];
@@ -201,12 +203,12 @@ typedef struct hv_input_translate_virtual_address {uint32_t padding;uint64_t control_flags;uint64_t gva_page; -} hv_input_translate_virtual_address; +}QEMU_PACKED hv_input_translate_virtual_address;typedef struct hv_output_translate_virtual_address {union hv_translate_gva_result translation_result;uint64_t gpa_page; -}hv_output_translate_virtual_address ; +} QEMU_PACKED hv_output_translate_virtual_address; typedef struct hv_register_x64_cpuid_result_parameters { struct {
@@ -226,13 +228,13 @@ typedef struct hv_register_x64_cpuid_result_parameters {uint32_t edx;uint32_t edx_mask; } result; -} hv_register_x64_cpuid_result_parameters; +}QEMU_PACKED hv_register_x64_cpuid_result_parameters;typedef struct hv_register_x64_msr_result_parameters {uint32_t msr_index;uint32_t access_type;uint32_t action;/* enum hv_unimplemented_msr_action */ -}hv_register_x64_msr_result_parameters ; +} QEMU_PACKED hv_register_x64_msr_result_parameters;union hv_register_intercept_result_parameters {struct hv_register_x64_cpuid_result_parameters cpuid;
@@ -244,6 +246,6 @@ typedef struct hv_input_register_intercept_result {uint32_t vp_index;uint32_t intercept_type;/* enum hv_intercept_type */ union hv_register_intercept_result_parameters parameters; -} hv_input_register_intercept_result; +}QEMU_PACKED hv_input_register_intercept_result;#endif /* HW_HYPERV_HVHDK_H */
Signature · 1 line
2.34.1
PATCH 3/10 accel/mshv: use mshv_create_partition_v2 payload
In reply to Magnus KulkeWhen using the extended request format we can instruct the hypervisor to provision enough space for requested XSAVE features. This is required for supporting QEMU models provided via the -cpu flag properly.
--- accel/mshv/mshv-all.c | 35 +++++-- include/hw/hyperv/hvgdk_mini.h | 2 + include/hw/hyperv/hvhdk.h | 185 +++++++++++++++++++++++++++++++++ 3 files changed, 214 insertions(+), 8 deletions(-)
accel/mshv/mshv-all.c +27 -8
@@ -112,19 +112,38 @@ static int resume_vm(int vm_fd)static int create_partition (int mshv_fd,int *vm_fd) { -int ret; -struct mshv_create_partition args = {0 }; +int ret, i; +uint64_t pt_flags; +union hv_partition_processor_xsave_features disabled_xsave_features; +union hv_partition_processor_features disabled_processor_features; +struct mshv_create_partition_v2 args = {0 };/* Initialize pt_flags with the desired features */ -uint64_t pt_flags = (1ULL <<MSHV_PT_BIT_LAPIC ) | - (1ULL <<MSHV_PT_BIT_X2APIC ) | - (1ULL <<MSHV_PT_BIT_GPA_SUPER_PAGES ); + pt_flags = (1ULL <<MSHV_PT_BIT_LAPIC ) | + (1ULL <<MSHV_PT_BIT_X2APIC ) | + (1ULL <<MSHV_PT_BIT_GPA_SUPER_PAGES ) | + (1ULL <<MSHV_PT_BIT_CPU_AND_XSAVE_FEATURES ); -/* Set default isolation type */ -uint64_t pt_isolation =MSHV_PT_ISOLATION_NONE ; +/* enable all */ + disabled_xsave_features.as_uint64 =0 ; +/* Enable all proc features by default */ +for (i =0 ; i <MSHV_NUM_CPU_FEATURES_BANKS ; i++) { + disabled_processor_features.as_uint64[i] =0 ; + } + disabled_processor_features.reserved_bank0 =1 ; +for (i =0 ; i <MSHV_NUM_CPU_FEATURES_BANKS ; i++) { + disabled_processor_features.as_uint64[i] =0 ; + } + +/* populate args structure */ args.pt_flags = pt_flags; - args.pt_isolation = pt_isolation; + args.pt_isolation =MSHV_PT_ISOLATION_NONE ; + args.pt_disabled_xsave = disabled_xsave_features.as_uint64; + args.pt_num_cpu_fbanks =MSHV_NUM_CPU_FEATURES_BANKS ; +for (i =0 ; i <MSHV_NUM_CPU_FEATURES_BANKS ; i++) { + args.pt_cpu_fbanks[i] = disabled_processor_features.as_uint64[i]; + } ret =ioctl (mshv_fd,MSHV_CREATE_PARTITION , &args);if (ret <0 ) {
include/hw/hyperv/hvgdk_mini.h +2 -0
@@ -456,6 +456,8 @@ typedef struct hv_input_set_vp_registers {struct hv_register_assoc elements[]; }QEMU_PACKED hv_input_set_vp_registers; +#define MSHV_VP_MAX_REGISTERS 128 +union hv_interrupt_control {uint64_t as_uint64;struct {
include/hw/hyperv/hvhdk.h +185 -0
191 lines not on this page · raw message
Signature · 1 line
2.34.1
Re: [PATCH 03/10] accel/mshv: use mshv_create_partition_v2 payload
Quoted 10 lines · Magnus Kulke
When using the extended request format we can instruct the hypervisor to provision enough space for requested XSAVE features. This is required for supporting QEMU models provided via the -cpu flag properly.
Signed-off-by Magnus Kulke <[email protected]>—
[…]
Hi,
On the 1st party Microsoft headers side (as in WHP), the two banks are defined separately.
For example:
ACountMCountSupport instead of a_count_m_count_support
Any reason in particular to diverge and would you prefer to use the definition here in WHP too?
Thank you,
Quoted 75 lines
include/hw/hyperv/hvhdk.h +68
-0+/* N.B. Begin bank 1 processor features. */ + uint64_t a_count_m_count_support:1; + uint64_t tsc_invariant_support:1; + uint64_t cl_zero_support:1; + uint64_t rdpru_support:1; + uint64_t la57_support:1; + uint64_t mbec_support:1; + uint64_t nested_virt_support:1; + uint64_t psfd_support:1; + uint64_t cet_ss_support:1; + uint64_t cet_ibt_support:1; + uint64_t vmx_exception_inject_support:1; + uint64_t enqcmd_support:1; + uint64_t umwait_tpause_support:1; + uint64_t movdiri_support:1; + uint64_t movdir64b_support:1; + uint64_t cldemote_support:1; + uint64_t serialize_support:1; + uint64_t tsc_deadline_tmr_support:1; + uint64_t tsc_adjust_support:1; + uint64_t fzl_rep_movsb:1; + uint64_t fs_rep_stosb:1; + uint64_t fs_rep_cmpsb:1; + uint64_t tsx_ld_trk_support:1; + uint64_t vmx_ins_outs_exit_info_support:1; + uint64_t hlat_support:1; + uint64_t sbdr_ssdp_no_support:1; + uint64_t fbsdp_no_support:1; + uint64_t psdp_no_support:1; + uint64_t fb_clear_support:1; + uint64_t btc_no_support:1; + uint64_t ibpb_rsb_flush_support:1; + uint64_t stibp_always_on_support:1; + uint64_t perf_global_ctrl_support:1; + uint64_t npt_execute_only_support:1; + uint64_t npt_ad_flags_support:1; + uint64_t npt1_gb_page_support:1; + uint64_t amd_processor_topology_node_id_support:1; + uint64_t local_machine_check_support:1; + uint64_t extended_topology_leaf_fp256_amd_support:1; + uint64_t gds_no_support:1; + uint64_t cmpccxadd_support:1; + uint64_t tsc_aux_virtualization_support:1; + uint64_t rmp_query_support:1; + uint64_t bhi_no_support:1; + uint64_t bhi_dis_support:1; + uint64_t prefetch_i_support:1; + uint64_t sha512_support:1; + uint64_t mitigation_ctrl_support:1; + uint64_t rfds_no_support:1; + uint64_t rfds_clear_support:1; + uint64_t sm3_support:1; + uint64_t sm4_support:1; + uint64_t secure_avic_support:1; + uint64_t guest_intercept_ctrl_support:1; + uint64_t sbpb_supported:1; + uint64_t ibpb_br_type_supported:1; + uint64_t srso_no_supported:1; + uint64_t srso_user_kernel_no_supported:1; + uint64_t vrew_clear_supported:1; + uint64_t tsa_l1_no_supported:1; + uint64_t tsa_sq_no_supported:1; + uint64_t lass_support:1; + uint64_t idle_hlt_intercept_support:1; + uint64_t msr_list_support:1; + }; +}; +enum
hv_translate_gva_result_code{
HV_TRANSLATE_GVA_SUCCESS= 0,Signature · 1 line
2.34.1
Re: [PATCH 03/10] accel/mshv: use mshv_create_partition_v2 payload
Quoted 10 lines · Mohamed Mediouni
On the 1st party Microsoft headers side (as in WHP), the two banks are defined separately.
For example:
ACountMCountSupport instead of
a_count_m_count_supportAny reason in particular to diverge and would you prefer to use the definition here in WHP too?
Thank you,
I am using the rust-vmm/mshv crates as a reference for the MSHV interface. I'm cherry picking the definition as we need them from this source usually:
I think in the future we want to move those definitions into the kernel's mshv UAPI header, so the WHPX code is probably more correct in this regard and the MSHV will do something similar in the future.
best,
magnus
Re: [PATCH 03/10] accel/mshv: use mshv_create_partition_v2 payload
In reply to Magnus KulkeQuoted 32 lines · Magnus Kulke
When using the extended request format we can instruct the hypervisor to provision enough space for requested XSAVE features. This is required for supporting QEMU models provided via the -cpu flag properly.
Signed-off-by Magnus Kulke <[email protected]>--- accel/mshv/mshv-all.c | 35 +++++-- include/hw/hyperv/hvgdk_mini.h | 2 + include/hw/hyperv/hvhdk.h | 185 +++++++++++++++++++++++++++++++++ 3 files changed, 214 insertions(+), 8 deletions(-)accel/mshv/mshv-all.c +6
-5@@ -112,19 +112,38 @@ static int resume_vm(int vm_fd)static int create_partition (int mshv_fd,int *vm_fd) { -int ret; -struct mshv_create_partition args = {0 }; +int ret, i; +uint64_t pt_flags; +union hv_partition_processor_xsave_features disabled_xsave_features; +union hv_partition_processor_features disabled_processor_features; +struct mshv_create_partition_v2 args = {0 };/* Initialize pt_flags with the desired features */ -uint64_t pt_flags = (1ULL <<MSHV_PT_BIT_LAPIC ) | - (1ULL <<MSHV_PT_BIT_X2APIC ) | - (1ULL <<MSHV_PT_BIT_GPA_SUPER_PAGES ); + pt_flags = (1ULL << MSHV_PT_BIT_LAPIC) |
accel/mshv/mshv-all.c static int resume_vm(int vm_fd) · line 130
Quoted 10 lines
accel/mshv/mshv-all.c +6
-2+ (1ULL <<MSHV_PT_BIT_X2APIC ) | + (1ULL <<MSHV_PT_BIT_GPA_SUPER_PAGES ) | + (1ULL <<MSHV_PT_BIT_CPU_AND_XSAVE_FEATURES ); -/* Set default isolation type */ -uint64_t pt_isolation =MSHV_PT_ISOLATION_NONE ; +/* enable all */ + disabled_xsave_features.as_uint64 =0 ; +/* Enable all proc features by default */
Instead of enabling all features by default, we could query which features are supported by the hypervisor on this host and enable only those.
To query supported features get the HvPartitionPropertyProcessorFeatures property via HvCallGetPartitionProperty (with HV_PARTITION_ID_SELF).
I think this would be more robust. This is what Cloud Hypervisor (well, technically, the mshv crate) does.
Thanks,
Anirudh.
Quoted 233 lines
accel/mshv/mshv-all.c +15
-1+for (i =0 ; i <MSHV_NUM_CPU_FEATURES_BANKS ; i++) { + disabled_processor_features.as_uint64[i] =0 ; + } + disabled_processor_features.reserved_bank0 =1 ; +for (i =0 ; i <MSHV_NUM_CPU_FEATURES_BANKS ; i++) { + disabled_processor_features.as_uint64[i] =0 ; + } + +/* populate args structure */ args.pt_flags = pt_flags; - args.pt_isolation = pt_isolation; + args.pt_isolation =MSHV_PT_ISOLATION_NONE ; + args.pt_disabled_xsave = disabled_xsave_features.as_uint64; + args.pt_num_cpu_fbanks =MSHV_NUM_CPU_FEATURES_BANKS ; +for (i =0 ; i <MSHV_NUM_CPU_FEATURES_BANKS ; i++) { + args.pt_cpu_fbanks[i] = disabled_processor_features.as_uint64[i]; + } ret =ioctl (mshv_fd,MSHV_CREATE_PARTITION , &args);if (ret <0 ) {include/hw/hyperv/hvgdk_mini.h +2
-0@@ -456,6 +456,8 @@ typedef struct hv_input_set_vp_registers {struct hv_register_assoc elements[]; }QEMU_PACKED hv_input_set_vp_registers; +#define MSHV_VP_MAX_REGISTERS 128 +union hv_interrupt_control {uint64_t as_uint64;struct {include/hw/hyperv/hvhdk.h +185
-0@@ -163,6 +163,191 @@ union hv_partition_synthetic_processor_features { }; }; +union hv_partition_processor_xsave_features { +struct { +uint64_t xsave_support:1 ; +uint64_t xsaveopt_support:1 ; +uint64_t avx_support:1 ; +uint64_t avx2_support:1 ; +uint64_t fma_support:1 ; +uint64_t mpx_support:1 ; +uint64_t avx512_support:1 ; +uint64_t avx512_dq_support:1 ; +uint64_t avx512_cd_support:1 ; +uint64_t avx512_bw_support:1 ; +uint64_t avx512_vl_support:1 ; +uint64_t xsave_comp_support:1 ; +uint64_t xsave_supervisor_support:1 ; +uint64_t xcr1_support:1 ; +uint64_t avx512_bitalg_support:1 ; +uint64_t avx512_i_fma_support:1 ; +uint64_t avx512_v_bmi_support:1 ; +uint64_t avx512_v_bmi2_support:1 ; +uint64_t avx512_vnni_support:1 ; +uint64_t gfni_support:1 ; +uint64_t vaes_support:1 ; +uint64_t avx512_v_popcntdq_support:1 ; +uint64_t vpclmulqdq_support:1 ; +uint64_t avx512_bf16_support:1 ; +uint64_t avx512_vp2_intersect_support:1 ; +uint64_t avx512_fp16_support:1 ; +uint64_t xfd_support:1 ; +uint64_t amx_tile_support:1 ; +uint64_t amx_bf16_support:1 ; +uint64_t amx_int8_support:1 ; +uint64_t avx_vnni_support:1 ; +uint64_t avx_ifma_support:1 ; +uint64_t avx_ne_convert_support:1 ; +uint64_t avx_vnni_int8_support:1 ; +uint64_t avx_vnni_int16_support:1 ; +uint64_t avx10_1_256_support:1 ; +uint64_t avx10_1_512_support:1 ; +uint64_t amx_fp16_support:1 ; +uint64_t reserved1:26 ; + }; +uint64_t as_uint64; +}; + +#define HV_PARTITION_PROCESSOR_FEATURES_BANKS 2 +#define HV_PARTITION_PROCESSOR_FEATURES_RESERVEDBANK1_BITFIELD_COUNT 4 + + +union hv_partition_processor_features { +uint64_t as_uint64[HV_PARTITION_PROCESSOR_FEATURES_BANKS ]; +struct { +uint64_t sse3_support:1 ; +uint64_t lahf_sahf_support:1 ; +uint64_t ssse3_support:1 ; +uint64_t sse4_1_support:1 ; +uint64_t sse4_2_support:1 ; +uint64_t sse4a_support:1 ; +uint64_t xop_support:1 ; +uint64_t pop_cnt_support:1 ; +uint64_t cmpxchg16b_support:1 ; +uint64_t altmovcr8_support:1 ; +uint64_t lzcnt_support:1 ; +uint64_t mis_align_sse_support:1 ; +uint64_t mmx_ext_support:1 ; +uint64_t amd3dnow_support:1 ; +uint64_t extended_amd3dnow_support:1 ; +uint64_t page_1gb_support:1 ; +uint64_t aes_support:1 ; +uint64_t pclmulqdq_support:1 ; +uint64_t pcid_support:1 ; +uint64_t fma4_support:1 ; +uint64_t f16c_support:1 ; +uint64_t rd_rand_support:1 ; +uint64_t rd_wr_fs_gs_support:1 ; +uint64_t smep_support:1 ; +uint64_t enhanced_fast_string_support:1 ; +uint64_t bmi1_support:1 ; +uint64_t bmi2_support:1 ; +uint64_t hle_support_deprecated:1 ; +uint64_t rtm_support_deprecated:1 ; +uint64_t movbe_support:1 ; +uint64_t npiep1_support:1 ; +uint64_t dep_x87_fpu_save_support:1 ; +uint64_t rd_seed_support:1 ; +uint64_t adx_support:1 ; +uint64_t intel_prefetch_support:1 ; +uint64_t smap_support:1 ; +uint64_t hle_support:1 ; +uint64_t rtm_support:1 ; +uint64_t rdtscp_support:1 ; +uint64_t clflushopt_support:1 ; +uint64_t clwb_support:1 ; +uint64_t sha_support:1 ; +uint64_t x87_pointers_saved_support:1 ; +uint64_t invpcid_support:1 ; +uint64_t ibrs_support:1 ; +uint64_t stibp_support:1 ; +uint64_t ibpb_support:1 ; +uint64_t unrestricted_guest_support:1 ; +uint64_t mdd_support:1 ; +uint64_t fast_short_rep_mov_support:1 ; +uint64_t l1dcache_flush_support:1 ; +uint64_t rdcl_no_support:1 ; +uint64_t ibrs_all_support:1 ; +uint64_t skip_l1df_support:1 ; +uint64_t ssb_no_support:1 ; +uint64_t rsb_a_no_support:1 ; +uint64_t virt_spec_ctrl_support:1 ; +uint64_t rd_pid_support:1 ; +uint64_t umip_support:1 ; +uint64_t mbs_no_support:1 ; +uint64_t mb_clear_support:1 ; +uint64_t taa_no_support:1 ; +uint64_t tsx_ctrl_support:1 ; +uint64_t reserved_bank0:1 ; + +/* N.B. Begin bank 1 processor features. */ +uint64_t a_count_m_count_support:1 ; +uint64_t tsc_invariant_support:1 ; +uint64_t cl_zero_support:1 ; +uint64_t rdpru_support:1 ; +uint64_t la57_support:1 ; +uint64_t mbec_support:1 ; +uint64_t nested_virt_support:1 ; +uint64_t psfd_support:1 ; +uint64_t cet_ss_support:1 ; +uint64_t cet_ibt_support:1 ; +uint64_t vmx_exception_inject_support:1 ; +uint64_t enqcmd_support:1 ; +uint64_t umwait_tpause_support:1 ; +uint64_t movdiri_support:1 ; +uint64_t movdir64b_support:1 ; +uint64_t cldemote_support:1 ; +uint64_t serialize_support:1 ; +uint64_t tsc_deadline_tmr_support:1 ; +uint64_t tsc_adjust_support:1 ; +uint64_t fzl_rep_movsb:1 ; +uint64_t fs_rep_stosb:1 ; +uint64_t fs_rep_cmpsb:1 ; +uint64_t tsx_ld_trk_support:1 ; +uint64_t vmx_ins_outs_exit_info_support:1 ; +uint64_t hlat_support:1 ; +uint64_t sbdr_ssdp_no_support:1 ; +uint64_t fbsdp_no_support:1 ; +uint64_t psdp_no_support:1 ; +uint64_t fb_clear_support:1 ; +uint64_t btc_no_support:1 ; +uint64_t ibpb_rsb_flush_support:1 ; +uint64_t stibp_always_on_support:1 ; +uint64_t perf_global_ctrl_support:1 ; +uint64_t npt_execute_only_support:1 ; +uint64_t npt_ad_flags_support:1 ; +uint64_t npt1_gb_page_support:1 ; +uint64_t amd_processor_topology_node_id_support:1 ; +uint64_t local_machine_check_support:1 ; +uint64_t extended_topology_leaf_fp256_amd_support:1 ; +uint64_t gds_no_support:1 ; +uint64_t cmpccxadd_support:1 ; +uint64_t tsc_aux_virtualization_support:1 ; +uint64_t rmp_query_support:1 ; +uint64_t bhi_no_support:1 ; +uint64_t bhi_dis_support:1 ; +uint64_t prefetch_i_support:1 ; +uint64_t sha512_support:1 ; +uint64_t mitigation_ctrl_support:1 ; +uint64_t rfds_no_support:1 ; +uint64_t rfds_clear_support:1 ; +uint64_t sm3_support:1 ; +uint64_t sm4_support:1 ; +uint64_t secure_avic_support:1 ; +uint64_t guest_intercept_ctrl_support:1 ; +uint64_t sbpb_supported:1 ; +uint64_t ibpb_br_type_supported:1 ; +uint64_t srso_no_supported:1 ; +uint64_t srso_user_kernel_no_supported:1 ; +uint64_t vrew_clear_supported:1 ; +uint64_t tsa_l1_no_supported:1 ; +uint64_t tsa_sq_no_supported:1 ; +uint64_t lass_support:1 ; +uint64_t idle_hlt_intercept_support:1 ; +uint64_t msr_list_support:1 ; + }; +}; +enum hv_translate_gva_result_code {HV_TRANSLATE_GVA_SUCCESS =0 ,Signature · 1 line
2.34.1
Re: [PATCH 03/10] accel/mshv: use mshv_create_partition_v2 payload
Quoted 10 lines
Instead of enabling all features by default, we could query which features are supported by the hypervisor on this host and enable only those.
To query supported features get the HvPartitionPropertyProcessorFeatures property via HvCallGetPartitionProperty (with
HV_PARTITION_ID_SELF).I think this would be more robust. This is what Cloud Hypervisor (well, technically, the mshv crate) does.
that makes sense, I'll adjust that part in a v2
thanks,
magnus
Re: [PATCH 03/10] accel/mshv: use mshv_create_partition_v2 payload
In reply to Anirudh RayabharamQuoted 48 lines · Anirudh Rayabharam
Quoted 32 lines · Magnus Kulke
When using the extended request format we can instruct the hypervisor to provision enough space for requested XSAVE features. This is required for supporting QEMU models provided via the -cpu flag properly.
Signed-off-by Magnus Kulke <[email protected]>--- accel/mshv/mshv-all.c | 35 +++++-- include/hw/hyperv/hvgdk_mini.h | 2 + include/hw/hyperv/hvhdk.h | 185 +++++++++++++++++++++++++++++++++ 3 files changed, 214 insertions(+), 8 deletions(-)accel/mshv/mshv-all.c +6
-5@@ -112,19 +112,38 @@ static int resume_vm(int vm_fd)static int create_partition (int mshv_fd,int *vm_fd) { -int ret; -struct mshv_create_partition args = {0 }; +int ret, i; +uint64_t pt_flags; +union hv_partition_processor_xsave_features disabled_xsave_features; +union hv_partition_processor_features disabled_processor_features; +struct mshv_create_partition_v2 args = {0 };/* Initialize pt_flags with the desired features */ -uint64_t pt_flags = (1ULL <<MSHV_PT_BIT_LAPIC ) | - (1ULL <<MSHV_PT_BIT_X2APIC ) | - (1ULL <<MSHV_PT_BIT_GPA_SUPER_PAGES ); + pt_flags = (1ULL << MSHV_PT_BIT_LAPIC) |accel/mshv/mshv-all.c static int resume_vm(int vm_fd) · line 130
Quoted 10 lines
accel/mshv/mshv-all.c +6
-2+ (1ULL <<MSHV_PT_BIT_X2APIC ) | + (1ULL <<MSHV_PT_BIT_GPA_SUPER_PAGES ) | + (1ULL <<MSHV_PT_BIT_CPU_AND_XSAVE_FEATURES ); -/* Set default isolation type */ -uint64_t pt_isolation =MSHV_PT_ISOLATION_NONE ; +/* enable all */ + disabled_xsave_features.as_uint64 =0 ; +/* Enable all proc features by default */ Instead of enabling all features by default, we could query which features are supported by the hypervisor on this host and enable only those.
Quoted 6 lines
To query supported features get the HvPartitionPropertyProcessorFeatures property via HvCallGetPartitionProperty (with
HV_PARTITION_ID_SELF).I think this would be more robust. This is what Cloud Hypervisor (well, technically, the mshv crate) does.
While I think querying the hypervisor is the correct action, are we sure this gives all the supported features on the host?
Is there a case that the Linux root partition has a reduced set of properties?
Wei
Re: [PATCH 03/10] accel/mshv: use mshv_create_partition_v2 payload
Quoted 52 lines · Wei Liu
Quoted 51 lines · Anirudh Rayabharam
Quoted 32 lines · Magnus Kulke
When using the extended request format we can instruct the hypervisor to provision enough space for requested XSAVE features. This is required for supporting QEMU models provided via the -cpu flag properly.
Signed-off-by Magnus Kulke <[email protected]>--- accel/mshv/mshv-all.c | 35 +++++-- include/hw/hyperv/hvgdk_mini.h | 2 + include/hw/hyperv/hvhdk.h | 185 +++++++++++++++++++++++++++++++++ 3 files changed, 214 insertions(+), 8 deletions(-)accel/mshv/mshv-all.c +6
-5@@ -112,19 +112,38 @@ static int resume_vm(int vm_fd)static int create_partition (int mshv_fd,int *vm_fd) { -int ret; -struct mshv_create_partition args = {0 }; +int ret, i; +uint64_t pt_flags; +union hv_partition_processor_xsave_features disabled_xsave_features; +union hv_partition_processor_features disabled_processor_features; +struct mshv_create_partition_v2 args = {0 };/* Initialize pt_flags with the desired features */ -uint64_t pt_flags = (1ULL <<MSHV_PT_BIT_LAPIC ) | - (1ULL <<MSHV_PT_BIT_X2APIC ) | - (1ULL <<MSHV_PT_BIT_GPA_SUPER_PAGES ); + pt_flags = (1ULL << MSHV_PT_BIT_LAPIC) |accel/mshv/mshv-all.c static int resume_vm(int vm_fd) · line 130
Quoted 10 lines
accel/mshv/mshv-all.c +6
-2+ (1ULL <<MSHV_PT_BIT_X2APIC ) | + (1ULL <<MSHV_PT_BIT_GPA_SUPER_PAGES ) | + (1ULL <<MSHV_PT_BIT_CPU_AND_XSAVE_FEATURES ); -/* Set default isolation type */ -uint64_t pt_isolation =MSHV_PT_ISOLATION_NONE ; +/* enable all */ + disabled_xsave_features.as_uint64 =0 ; +/* Enable all proc features by default */ Instead of enabling all features by default, we could query which features are supported by the hypervisor on this host and enable only those.
To query supported features get the HvPartitionPropertyProcessorFeatures property via HvCallGetPartitionProperty (with
HV_PARTITION_ID_SELF).
Quoted 6 lines
Quoted 3 lines
I think this would be more robust. This is what Cloud Hypervisor (well, technically, the mshv crate) does.
While I think querying the hypervisor is the correct action, are we sure this gives all the supported features on the host?
Yes, I confirmed this with the hypervisor team when implementing it in the mshv crate.
Quoted 3 lines
Is there a case that the Linux root partition has a reduced set of properties?
It is my understanding that the processor features we expose to a guest must always be a subset of the processor features of the root/parent partition. So, if some proc feature is not available for root, we cannot make it available for any guest partitions either.
Thanks,
Anirudh.
Quoted 3 lines
Wei
Re: [PATCH 03/10] accel/mshv: use mshv_create_partition_v2 payload
On Thu, Feb 19, 2026 at 10:33:44AM +0000, Anirudh Rayabharam wrote:
[...]
Quoted 19 lines
Quoted 8 lines
Quoted 8 lines
Instead of enabling all features by default, we could query which features are supported by the hypervisor on this host and enable only those.
To query supported features get the HvPartitionPropertyProcessorFeatures property via HvCallGetPartitionProperty (with
HV_PARTITION_ID_SELF).Quoted 6 lines
Quoted 3 lines
I think this would be more robust. This is what Cloud Hypervisor (well, technically, the mshv crate) does.
While I think querying the hypervisor is the correct action, are we sure this gives all the supported features on the host?
Yes, I confirmed this with the hypervisor team when implementing it in the mshv crate.
Quoted 1 line
Quoted 7 lines
Quoted 2 lines
Is there a case that the Linux root partition has a reduced set of properties?
It is my understanding that the processor features we expose to a guest must always be a subset of the processor features of the root/parent partition. So, if some proc feature is not available for root, we cannot make it available for any guest partitions either.
One thing comes to mind is the confidential VM support. In our CVM implementation, the guest can see a different set of features than the root partition, mostly around SEV-SNP support. I cannot say 100% for sure that if this matters.
Wei
Quoted 7 lines
Thanks,
Anirudh.Quoted 3 lines
Wei
PATCH 4/10 target/i386/mshv: fix cpuid propagation bug
In reply to Magnus KulkeThe list argument was missing a pointer and hence e.g. -cpu qemu64 and -cpu EPYC would produce the same of cpuid leaves.
--- target/i386/mshv/mshv-cpu.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
target/i386/mshv/mshv-cpu.c +4 -4
@@ -441,7 +441,7 @@ int mshv_load_regs(CPUState *cpu)return 0 ; } -static void add_cpuid_entry (GList *cpuid_entries, +static void add_cpuid_entry (GList **cpuid_entries,uint32_t function,uint32_t index,uint32_t eax,uint32_t ebx,uint32_t ecx,uint32_t edx)
@@ -456,10 +456,10 @@ static void add_cpuid_entry(GList *cpuid_entries, entry->ecx = ecx; entry->edx = edx; - cpuid_entries =g_list_append (cpuid_entries, entry); + *cpuid_entries =g_list_append (*cpuid_entries, entry); } -static void collect_cpuid_entries (const CPUState *cpu,GList *cpuid_entries) +static void collect_cpuid_entries (const CPUState *cpu,GList **cpuid_entries) {X86CPU *x86_cpu =X86_CPU (cpu);CPUX86State *env = &x86_cpu->env;
@@ -615,7 +615,7 @@ static int set_cpuid2(const CPUState *cpu)struct hv_cpuid_entry *entry;GList *entries =NULL ; -collect_cpuid_entries (cpu, entries); +collect_cpuid_entries (cpu, &entries); n_entries =g_list_length (entries); cpuid_size =sizeof (struct hv_cpuid )
Signature · 1 line
2.34.1
Re: [PATCH 04/10] target/i386/mshv: fix cpuid propagation bug
Quoted 4 lines · Magnus Kulke
The list argument was missing a pointer and hence e.g. -cpu qemu64 and -cpu EPYC would produce the same of cpuid leaves.
Signed-off-by Magnus Kulke <[email protected]>
PATCH 5/10 target/i386/mshv: fix various cpuid traversal bugs
In reply to Magnus Kulke- Hardcoded max_leaf was not accurate. We query leaf 0x0 and 0x80000000 to get the actual max leaves
- On all 0 zeroes on leaf 0x0d, we register 0-63 subleaves with zeros indicating XSAVE is disabled
- Subleaf 0 was hardcoded, so the Hypervisor returned defaults for other subleaves
- Subleaf 0 war hardcoded, so we were passing 0 instead of actual subleaf when adding entries. We now pass the correct subleaf value to add_cpuid_entry()
- Leaves 0x04,0x07,0d,0f,10 weren't marked as subleaf-specific
--- target/i386/mshv/mshv-cpu.c | 88 +++++++++++++++++++++++++------------ 1 file changed, 61 insertions(+), 27 deletions(-)
target/i386/mshv/mshv-cpu.c +61 -27
@@ -465,14 +465,17 @@ static void collect_cpuid_entries(const CPUState *cpu, GList **cpuid_entries)CPUX86State *env = &x86_cpu->env;uint32_t eax, ebx, ecx, edx;uint32_t leaf, subleaf; -size_t max_leaf =0x1F ; -size_t max_subleaf =0x20 ; - -uint32_t leaves_with_subleaves[] = {0x4 ,0x7 ,0xD ,0xF ,0x10 }; +uint32_t max_basic_leaf, max_extended_leaf; +uint32_t max_subleaf =0x20 ; +uint32_t leaves_with_subleaves[] = {0x04 ,0x07 ,0x0d ,0x0f ,0x10 };int n_subleaf_leaves =ARRAY_SIZE (leaves_with_subleaves); -/* Regular leaves without subleaves */ -for (leaf =0 ; leaf <= max_leaf; leaf++) { +/* Get maximum basic and and extended CPUID leaves */ +cpu_x86_cpuid (env,0 ,0 , &max_basic_leaf, &ebx, &ecx, &edx); +cpu_x86_cpuid (env,0x80000000 ,0 , &max_extended_leaf, &ebx, &ecx, &edx); + +/* Collect basic leaves (0x0 to max_basic_leaf) */ +for (leaf =0 ; leaf <= max_basic_leaf; leaf++) {bool has_subleaves = false;for (int i =0 ; i < n_subleaf_leaves; i++) {if (leaf == leaves_with_subleaves[i]) {
@@ -483,27 +486,40 @@ static void collect_cpuid_entries(const CPUState *cpu, GList **cpuid_entries)if (!has_subleaves) {cpu_x86_cpuid (env, leaf,0 , &eax, &ebx, &ecx, &edx); -if (eax ==0 && ebx ==0 && ecx ==0 && edx ==0 ) { -/* all zeroes indicates no more leaves */ -continue ; - } -add_cpuid_entry (cpuid_entries, leaf,0 , eax, ebx, ecx, edx);continue ; } +/* +* Valid XSAVE components can exist at a higher index se we need to set +* all subleaves for leaf 0x0d, even if we encounter an empty one. +*/ +if (leaf ==0x0d ) { +for (subleaf =0 ; subleaf <=63 ; subleaf++) { +cpu_x86_cpuid (env, leaf, subleaf, &eax, &ebx, &ecx, &edx); +add_cpuid_entry (cpuid_entries, leaf, subleaf, + eax, ebx, ecx, edx); + } +continue ; + } + subleaf =0 ;while (subleaf < max_subleaf) {cpu_x86_cpuid (env, leaf, subleaf, &eax, &ebx, &ecx, &edx);if (eax ==0 && ebx ==0 && ecx ==0 && edx ==0 ) { -/* all zeroes indicates no more leaves */ break ; } -add_cpuid_entry (cpuid_entries, leaf,0 , eax, ebx, ecx, edx); +add_cpuid_entry (cpuid_entries, leaf, subleaf, eax, ebx, ecx, edx); subleaf++; } } + +/* Collect extended leaves (0x80000000 to max_extended_leaf) */ +for (leaf =0x80000000 ; leaf <= max_extended_leaf; leaf++) { +cpu_x86_cpuid (env, leaf,0 , &eax, &ebx, &ecx, &edx); +add_cpuid_entry (cpuid_entries, leaf,0 , eax, ebx, ecx, edx); + } } static int register_intercept_result_cpuid_entry(const CPUState *cpu,
@@ -576,22 +592,40 @@ static int register_intercept_result_cpuid(const CPUState *cpu, subleaf_specific =0 ; always_override =1 ; -/* Intel */ -/* 0xb - Extended Topology Enumeration Leaf */ -/* 0x1f - V2 Extended Topology Enumeration Leaf */ -/* AMD */ -/* 0x8000_001e - Processor Topology Information */ -/* 0x8000_0026 - Extended CPU Topology */ -if (entry->function ==0xb - || entry->function ==0x1f - || entry->function ==0x8000001e - || entry->function ==0x80000026 ) { +/* +* Intel +* 0xb - Extended Topology Enumeration Leaf +* 0x1f - V2 Extended Topology Enumeration Leaf +* AMD +* 0x8000_001e - Processor Topology Information +* 0x8000_0026 - Extended CPU Topology +*/ +if (entry->function ==0xb || + entry->function ==0x1f || + entry->function ==0x8000001e || + entry->function ==0x80000026 ) { + subleaf_specific =1 ; + always_override =1 ; +/* +* Feature enumeration leaves (subleaf-specific) +* 0x04: Deterministic Cache Parameters +* 0x07: Structured Extended Feature Flags +* 0x0D: Processor Extended State Enumeration +* 0x0F: Platform QoS Monitoring +* 0x10: Platform QoS Enforcement +*/ + }else if (entry->function ==0x04 || + entry->function ==0x07 || + entry->function ==0x0d || + entry->function ==0x0f || + entry->function ==0x10 ) { subleaf_specific =1 ; always_override =1 ; - }else if (entry->function ==0x00000001 - || entry->function ==0x80000000 - || entry->function ==0x80000001 - || entry->function ==0x80000008 ) { +/* Basic feature leaves (no subleaves) */ + }else if (entry->function ==0x00000001 || + entry->function ==0x80000000 || + entry->function ==0x80000001 || + entry->function ==0x80000008 ) { subleaf_specific =0 ; always_override =1 ; }
Signature · 1 line
2.34.1
PATCH 6/10 target/i386/mshv: change cpuid mask to UINT32_MAX
In reply to Magnus KulkeThe current implementation would only override the feature bits that are enabled, however we also want to consider disabled features, hence all bits are set on the masks in the hypercall argument.
--- target/i386/mshv/mshv-cpu.c | 17 ++++++----------- 1 file changed, 6 insertions(+), 11 deletions(-)
target/i386/mshv/mshv-cpu.c +6 -11
@@ -538,22 +538,17 @@ static int register_intercept_result_cpuid_entry(const CPUState *cpu, .input.always_override = always_override, .input.padding = 0,/* -* With regard to masks - these are to specify bits to be overwritten -* The current CpuidEntry structure wouldn't allow to carry the masks -* in addition to the actual register values. For this reason, the -* masks are set to the exact values of the corresponding register bits -* to be registered for an overwrite. To view resulting values the -* hypervisor would return, HvCallGetVpCpuidValues hypercall can be -* used. +* Masks specify which bits to override. Set to 0xFFFFFFFF to +* override all bits with the values from the QEMU CPU model. */ .result.eax = entry->eax, - .result.eax_mask = entry->eax, + .result.eax_mask = 0xFFFFFFFF, .result.ebx = entry->ebx, - .result.ebx_mask = entry->ebx, + .result.ebx_mask = 0xFFFFFFFF, .result.ecx = entry->ecx, - .result.ecx_mask = entry->ecx, + .result.ecx_mask = 0xFFFFFFFF, .result.edx = entry->edx, - .result.edx_mask = entry->edx, + .result.edx_mask = 0xFFFFFFFF, };union hv_register_intercept_result_parameters parameters = { .cpuid = cpuid_params,
Signature · 1 line
2.34.1
PATCH 7/10 target/i386/mshv: set cpu model name on -cpu host
In reply to Magnus KulkeBy registering a cpu_accel_class for mshv, we can report an accurate model name on -cpu host.
--- target/i386/mshv/mshv-cpu.c | 31 +++++++++++++++++++++++++++++++ 1 file changed, 31 insertions(+)
target/i386/mshv/mshv-cpu.c +31 -0
@@ -24,10 +24,13 @@#include "hw/i386/apic_internal.h" #include "cpu.h" +#include "host-cpu.h" #include "emulate/x86_decode.h" #include "emulate/x86_emu.h" #include "emulate/x86_flags.h" +#include "accel/accel-cpu-target.h" +#include "trace-accel_mshv.h" #include "trace.h"
@@ -1793,3 +1796,31 @@ int mshv_arch_post_init_vm(int vm_fd)return ret; } + +static void mshv_cpu_instance_init (CPUState *cs) +{ +X86CPU *cpu =X86_CPU (cs); + +host_cpu_instance_init (cpu); +} + +static void mshv_cpu_accel_class_init (ObjectClass *oc,const void *data) +{ +AccelCPUClass *acc =ACCEL_CPU_CLASS (oc); + + acc->cpu_instance_init = mshv_cpu_instance_init; +} + +static const TypeInfo mshv_cpu_accel_type_info = { + .name =ACCEL_CPU_NAME ("mshv" ), + .parent =TYPE_ACCEL_CPU , + .class_init = mshv_cpu_accel_class_init, + .abstract = true, +}; + +static void mshv_cpu_accel_register_types (void ) +{ +type_register_static (&mshv_cpu_accel_type_info); +} + +type_init (mshv_cpu_accel_register_types);
Signature · 1 line
2.34.1
PATCH 8/10 target/i386: query mshv accel for supported cpuids
In reply to Magnus KulkeWe implement mshv_get_supported_cpuid() and invoke it in x86_cpu_get_supported_feature_word() retrieve the cpu features that the host is supporting. Initially we mask the virtualization capabilitities potentially we might need to mask more in the future.
--- include/system/mshv.h | 3 +++ target/i386/cpu.c | 8 ++++++++ target/i386/mshv/mshv-cpu.c | 27 +++++++++++++++++++++++++++ 3 files changed, 38 insertions(+)
include/system/mshv.h +3 -0
@@ -60,4 +60,7 @@ int mshv_irqchip_add_irqfd_notifier_gsi(const EventNotifier *n, const EventNotifier *rn, int virq);int mshv_irqchip_remove_irqfd_notifier_gsi (const EventNotifier *n,int virq); +/* cpuid */ +uint32_t mshv_get_supported_cpuid (uint32_t func,uint32_t idx,int reg); +#endif
target/i386/cpu.c +8 -0
@@ -26,6 +26,7 @@#include "tcg/helper-tcg.h" #include "exec/translation-block.h" #include "system/hvf.h" +#include "system/mshv.h" #include "hvf/hvf-i386.h" #include "kvm/kvm_i386.h" #include "kvm/tdx.h"
@@ -7958,6 +7959,13 @@ uint64_t x86_cpu_get_supported_feature_word(X86CPU *cpu, FeatureWord w) r =hvf_get_supported_cpuid (wi->cpuid.eax, wi->cpuid.ecx, wi->cpuid.reg); + }else if (mshv_enabled ()) { +if (wi->type !=CPUID_FEATURE_WORD ) { +return 0 ; + } + r =mshv_get_supported_cpuid (wi->cpuid.eax, + wi->cpuid.ecx, + wi->cpuid.reg); }else if (tcg_enabled ()) { r = wi->tcg_features; }else {
target/i386/mshv/mshv-cpu.c +27 -0
@@ -1757,6 +1757,33 @@ void mshv_arch_destroy_vcpu(CPUState *cpu)g_clear_pointer (&env->emu_mmio_buf, g_free); } +uint32_t mshv_get_supported_cpuid (uint32_t func,uint32_t idx,int reg) +{ +uint32_t eax, ebx, ecx, edx; +uint32_t ret =0 ; + +host_cpuid (func, idx, &eax, &ebx, &ecx, &edx); +switch (reg) { +case R_EAX : + ret = eax;break ; +case R_EBX : + ret = ebx;break ; +case R_ECX : + ret = ecx;break ; +case R_EDX : + ret = edx;break ; + } + +/* Disable nested virtualization features not yet supported by MSHV */ +if (func ==0x80000001 && reg ==R_ECX ) { + ret &= ~CPUID_EXT3_SVM ; + } +if (func ==0x01 && reg ==R_ECX ) { + ret &= ~CPUID_EXT_VMX ; + } +return ret; +} +/* * Default Microsoft Hypervisor behavior for unimplemented MSR is to send a * fault to the guest if it tries to access it. It is possible to override
Signature · 1 line
2.34.1
Re: [PATCH 08/10] target/i386: query mshv accel for supported cpuids
Quoted 10 lines · Magnus Kulke
We implement
mshv_get_supported_cpuid()and invoke it inx86_cpu_get_supported_feature_word()retrieve the cpu features that the host is supporting. Initially we mask the virtualization capabilitities potentially we might need to mask more in the future.Signed-off-by Magnus Kulke <[email protected]>
Hi,
Interesting. For WHPX on Arm, what’s done is creating a stub partition and extract CPUID info from it (with necessary guest stub code if necessary) but doing it this way like HVF x86 does is intriguing too…
Is this future-proof or are there catches with this approach?
Thank you,
Quoted 86 lines
---
include/system/mshv.h | 3 +++ target/i386/cpu.c | 8 ++++++++ target/i386/mshv/mshv-cpu.c | 27 +++++++++++++++++++++++++++ 3 files changed, 38 insertions(+)include/system/mshv.h +3
-0@@ -60,4 +60,7 @@ int mshv_irqchip_add_irqfd_notifier_gsi(const EventNotifier *n, const EventNotifier *rn, int virq);int mshv_irqchip_remove_irqfd_notifier_gsi (const EventNotifier *n,int virq); +/* cpuid */ +uint32_t mshv_get_supported_cpuid (uint32_t func,uint32_t idx,int reg); +#endif target/i386/cpu.c +8
-0@@ -26,6 +26,7 @@#include "tcg/helper-tcg.h" #include "exec/translation-block.h" #include "system/hvf.h" +#include "system/mshv.h" #include "hvf/hvf-i386.h" #include "kvm/kvm_i386.h" #include "kvm/tdx.h" @@ -7958,6 +7959,13 @@ uint64_t x86_cpu_get_supported_feature_word(X86CPU *cpu, FeatureWord w) r =hvf_get_supported_cpuid (wi->cpuid.eax, wi->cpuid.ecx, wi->cpuid.reg); + }else if (mshv_enabled ()) { +if (wi->type !=CPUID_FEATURE_WORD ) { +return 0 ; + } + r =mshv_get_supported_cpuid (wi->cpuid.eax, + wi->cpuid.ecx, + wi->cpuid.reg); }else if (tcg_enabled ()) { r = wi->tcg_features; }else {target/i386/mshv/mshv-cpu.c +27
-0@@ -1757,6 +1757,33 @@ void mshv_arch_destroy_vcpu(CPUState *cpu)g_clear_pointer (&env->emu_mmio_buf, g_free); } +uint32_t mshv_get_supported_cpuid (uint32_t func,uint32_t idx,int reg) +{ +uint32_t eax, ebx, ecx, edx; +uint32_t ret =0 ; + +host_cpuid (func, idx, &eax, &ebx, &ecx, &edx); +switch (reg) { +case R_EAX : + ret = eax;break ; +case R_EBX : + ret = ebx;break ; +case R_ECX : + ret = ecx;break ; +case R_EDX : + ret = edx;break ; + } + +/* Disable nested virtualization features not yet supported by MSHV */ +if (func ==0x80000001 && reg ==R_ECX ) { + ret &= ~CPUID_EXT3_SVM ; + } +if (func ==0x01 && reg ==R_ECX ) { + ret &= ~CPUID_EXT_VMX ; + } +return ret; +} +/* * Default Microsoft Hypervisor behavior for unimplemented MSR is to send a * fault to the guest if it tries to access it. It is possible to override Signature · 1 line
2.34.1
Re: [PATCH 08/10] target/i386: query mshv accel for supported cpuids
Quoted 6 lines · Mohamed Mediouni
Interesting. For WHPX on Arm, what’s done is creating a stub partition and extract CPUID info from it (with necessary guest stub code if necessary) but doing it this way like HVF x86 does is intriguing too…
Is this future-proof or are there catches with this approach?
I think this is a valid approach that we also considered. I assume for MSHV eventually we want to have a MSHV_GET_SUPPORTED_CPUID ioctl, similar to what exists in KVM, so we can make centralize some decisions instead of leaving it up to the VMM. This approach might end up using a dummy partition inside the kernel, but would be an implementation detail.
Re: [PATCH 08/10] target/i386: query mshv accel for supported cpuids
In reply to Mohamed MediouniQuoted 12 lines · Mohamed Mediouni
Quoted 11 lines
On 11. Feb 2026, at 16:54, Magnus Kulke
<[email protected]> wrote:We implement
mshv_get_supported_cpuid()and invoke it inx86_cpu_get_supported_feature_word()retrieve the cpu features that the host is supporting. Initially we mask the virtualization
capabilitities potentially we might need to mask more in the
future.Signed-off-by Magnus Kulke <[email protected]>
Quoted 8 lines
Hi,
Interesting. For WHPX on Arm, what’s done is creating a stub partition and extract CPUID info from it (with necessary guest stub code if necessary) but doing it this way like HVF x86 does is intriguing too…
Is this future-proof or are there catches with this approach?
No, it's absolutely not future proof.
The difference is two-fold: first, HVF on x86 essentially implements the hypervisor itself. It operates at the VMCS level and is a mix of KVM and "regular" QEMU code. Second, for feature leaves HVF lists explicitly all the supported bits and it also limits EAX to 0xD (it does not try too hard with AMD leaves at 0x800000nn because HVF x86 is Intel-only anyway).
So, what this patch implements is not like HVF x86. The stub partition idea is nice and more future proof.
Paolo
PATCH 9/10 target/i386/mshv: populate xsave area offsets
In reply to Magnus KulkeThis function is copied verbatim from kvm_cpu_xsave_init(), for MSHV we need to populate the offsets similarly.
--- target/i386/mshv/mshv-cpu.c | 31 +++++++++++++++++++++++++++++++ 1 file changed, 31 insertions(+)
target/i386/mshv/mshv-cpu.c +31 -0
@@ -1824,11 +1824,42 @@ int mshv_arch_post_init_vm(int vm_fd)return ret; } +static void mshv_cpu_xsave_init (void ) +{ +static bool first = true; +uint32_t eax, ebx, ecx, edx; +int i; + +if (!first) { +return ; + } + first = false; + +/* x87 and SSE states are in the legacy region of the XSAVE area. */ + x86_ext_save_areas[XSTATE_FP_BIT ].offset =0 ; + x86_ext_save_areas[XSTATE_SSE_BIT ].offset =0 ; + +for (i =XSTATE_SSE_BIT +1 ; i <XSAVE_STATE_AREA_COUNT ; i++) { +ExtSaveArea *esa = &x86_ext_save_areas[i]; + +if (!esa->size) { +continue ; + } +host_cpuid (0xd , i, &eax, &ebx, &ecx, &edx); +if (eax !=0 ) { +assert (esa->size == eax); + esa->offset = ebx; + esa->ecx = ecx; + } + } +} +static void mshv_cpu_instance_init (CPUState *cs) {X86CPU *cpu =X86_CPU (cs);host_cpu_instance_init (cpu); +mshv_cpu_xsave_init (); }static void mshv_cpu_accel_class_init (ObjectClass *oc,const void *data)
Signature · 1 line
2.34.1
PATCH 10/10 target/i386/mshv: filter out CET bits in cpuid
In reply to Magnus Kulkee46dcef1d0 introduced CET_U/CET_S xstate bits in the ExtSaveArea. The response in EBX of a 0x4,1 cpuid query will contain a size contingent on what the guest enabled in the IA32_XSS MSR (compacted xsave size).
Since the MSHV accelerator currently responds with static answers to 0x4d cpuid queries, we'll have to disable those bits, even if the host supports them.
--- target/i386/mshv/mshv-cpu.c | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+)
target/i386/mshv/mshv-cpu.c +19 -0
@@ -1781,6 +1781,25 @@ uint32_t mshv_get_supported_cpuid(uint32_t func, uint32_t idx, int reg)if (func ==0x01 && reg ==R_ECX ) { ret &= ~CPUID_EXT_VMX ; } + +/* +* MSHV currently uses static CPUID intercept results for leaf 0xD. +* However there are feature-responses that are dynamic based on what a +* guest enables in XCR0 and XSS, such as CET shadow stack. +* +* A guest which doesn't know about those features yet would encounter an +* unexpcted CPUID[0xD,1].EBX (compactes XSAVE size) and either fail +* or gracefully degrade by not using XSAVE at all. +* +* To avoid this, we filter out supervisor xstate features. +*/ +if (func ==0x07 && idx ==0 && reg ==R_ECX ) { + ret &= ~CPUID_7_0_ECX_CET_SHSTK ; + } +if (func ==0x07 && idx ==0 && reg ==R_EDX ) { + ret &= ~CPUID_7_0_EDX_CET_IBT ; + } +return ret; }
Signature · 1 line
2.34.1
Quoted 13 lines · Magnus Kulke
Hey all,
In the current MSHV accelerator code passing CPU features via the -cpu flag doesn't work as intended yet. When using the MSHV hypervisor we either silently discard the specified model/features and leave it up to the hypervisor to provide a sensible set of features or if the user selects -cpu host, the hypervisor might refuse to create a partition.
This changeset introduces a more comprehensive support for passing desired guest cpu features to the hypervisor. It's also a prerequisite for Live Migration support, in which we have to roundtrip CPU State explicitly
Quoted 6 lines
Known issues:
We will probably have to iterate a bit more on this, since recently support for CET_U/CET_S xstate bits has been introduced in QEMU, which doesn't harmonize with our current approach of configuring the hypervisor with static responses to cpuid queries.
Hi Magnus,
I went back to reviewing this series.
More or less, using "-cpu host" should work because MSHV already runs in a partition. Therefore, it should be safe to assume that whatever bits were allowed in the current partition's CPUID will also be allowed in the nested guest.
However, you still need to mask the features corresponding to MSRs that you do not save/restore; this includes for example TSC deadline timer, AMX (XFD), FRED, PMU, UMWAIT are the first few that came to mind. Or alternatively, just add them to get/put_msrs.
Later on, we probably want to share some of the code to handle MSRs between Hyper-V and KVM. Please add some comments about the hypercalls, since they are poorly documented, explaining how to find out which MSRs are supported by the hypervisor.
With respect to live migration, here are a few bits of states that should be migrated:
- the FPU registers (MshvFPU is dead code and can be removed; I missed it during my initial review).
- the PDPTRs. That is probably okay (because you never set nor read env->pdptrs_valid) but please check if Hyper-V supports reading them.
- KVM also has support for migrating in the middle of an exception being delivered (for example if an EPT violation happens due to a write to the stack); this is handled with fields such as these:
VMSTATE_UINT8(env.exception_pending, X86CPU),
VMSTATE_UINT8(env.exception_injected, X86CPU),
VMSTATE_UINT8(env.exception_has_payload, X86CPU),
VMSTATE_UINT64(env.exception_payload, X86CPU),
VMSTATE_INT32(env.exception_nr, X86CPU),
VMSTATE_INT32(env.interrupt_injected, X86CPU),
VMSTATE_UINT8(env.soft_interrupt, X86CPU),
VMSTATE_UINT8(env.nmi_injected, X86CPU),
VMSTATE_UINT8(env.nmi_pending, X86CPU),please check how Hyper-V handles this situation so that it can be implemented in QEMU as well.
Thanks,
Paolo
Quoted 9 lines · Paolo Bonzini
Hi Magnus,
I went back to reviewing this series.
More or less, using "-cpu host" should work because MSHV already runs in a partition. Therefore, it should be safe to assume that whatever bits were allowed in the current partition's CPUID will also be allowed in the nested guest.
Quoted 5 lines
However, you still need to mask the features corresponding to MSRs that you do not save/restore; this includes for example TSC deadline timer, AMX (XFD), FRED, PMU, UMWAIT are the first few that came to mind. Or alternatively, just add them to get/put_msrs.
Hey Paolo,
thanks for taking a look at this. I am currently staging follow up patch-sets in which the MSR handling is reworked as part of the live migration support. In those there will be more MSRs that are being covered in a migration, using a hardcoded list in rust-vmm/mshv as base, with some additional ones added:
https://docs.rs/crate/mshv-ioctls/0.6.7/source/src/ioctls/system.rs#363
in the future I think we want to have a MSHV_GET_SUPPORTED_MSRS ioctl that we can query, similar to what's available for KVM.
There are "hv_partition_processor_features" that we query from the hypervisor to filter out MSRs that are not available for a given partition, e.g.
uint64_t tsc_adjust_support:1;
Quoted 5 lines
Later on, we probably want to share some of the code to handle MSRs between Hyper-V and KVM. Please add some comments about the hypercalls, since they are poorly documented, explaining how to find out which MSRs are supported by the hypervisor.
If the hypercalls are not documented, we probably want to fix it either in the rust-vmm/mshv create (which at the moment provides authorative headers (until we have moved everything to the mshv uapi headers), but i'll double check.
Quoted 5 lines
With respect to live migration, here are a few bits of states that should be migrated:
- the FPU registers (MshvFPU is dead code and can be removed; I missed it during my initial review).
yup, the FPU registers handling has also been reworked a bit to accomodate XSAVE migration.
Quoted 4 lines
- the PDPTRs. That is probably okay (because you never set nor read
env->pdptrs_valid) but please check if Hyper-V supports reading them.
I'll try to find that out. Haven't stumbled over this so far, but I understand it's relevant for 32bit PAE guest, which we probably haven't tested thoroughly yet.
Quoted 14 lines
- KVM also has support for migrating in the middle of an exception being delivered (for example if an EPT violation happens due to a write to the stack); this is handled with fields such as these:
VMSTATE_UINT8(env.exception_pending, X86CPU), VMSTATE_UINT8(env.exception_injected, X86CPU), VMSTATE_UINT8(env.exception_has_payload, X86CPU), VMSTATE_UINT64(env.exception_payload, X86CPU), VMSTATE_INT32(env.exception_nr, X86CPU), VMSTATE_INT32(env.interrupt_injected, X86CPU), VMSTATE_UINT8(env.soft_interrupt, X86CPU), VMSTATE_UINT8(env.nmi_injected, X86CPU), VMSTATE_UINT8(env.nmi_pending, X86CPU),
Quoted 3 lines
please check how Hyper-V handles this situation so that it can be implemented in QEMU as well.
Those are covered in MSHV's "vCPU Events", I think they map quite cleanly to the QEMU representation:
https://docs.rs/mshv-bindings/0.6.7/src/mshv_bindings/x86_64/regs.rs.html#404
thanks,
magnus
Quoted 17 lines · Magnus Kulke
Quoted 9 lines · Paolo Bonzini
Hi Magnus,
I went back to reviewing this series.
More or less, using "-cpu host" should work because MSHV already runs in a partition. Therefore, it should be safe to assume that whatever bits were allowed in the current partition's CPUID will also be allowed in the nested guest.
Quoted 5 lines
However, you still need to mask the features corresponding to MSRs that you do not save/restore; this includes for example TSC deadline timer, AMX (XFD), FRED, PMU, UMWAIT are the first few that came to mind. Or alternatively, just add them to get/put_msrs.
Quoted 5 lines
Hey Paolo,
thanks for taking a look at this. I am currently staging follow up patch-sets in which the MSR handling is reworked as part of the live migration support.
Ok, let's start from there so that we don't have to think about filtering out features because they're not migrated.
Thanks for the extra information, I'll study the rust-vmm docs.
Quoted 2 lines
in the future I think we want to have a
MSHV_GET_SUPPORTED_MSRSioctl that we can query, similar to what's available for KVM.
That may not be needed if we can rely on the parent partition's CPUID (I changed my mind and I think we can, because QEMU already filters out CPUID features it doesn't know about).
We may want to share the knowledge of MSRs between KVM and MSHV, but that's a different story.
Paolo
Quoted 62 lines
There are "
hv_partition_processor_features" that we query from the hypervisor to filter out MSRs that are not available for a given partition, e.g.uint64_t tsc_adjust_support:1;Quoted 5 lines
Later on, we probably want to share some of the code to handle MSRs between Hyper-V and KVM. Please add some comments about the hypercalls, since they are poorly documented, explaining how to find out which MSRs are supported by the hypervisor.
If the hypercalls are not documented, we probably want to fix it either in the rust-vmm/mshv create (which at the moment provides authorative headers (until we have moved everything to the mshv uapi headers), but i'll double check.
Quoted 5 lines
With respect to live migration, here are a few bits of states that should be migrated:
- the FPU registers (MshvFPU is dead code and can be removed; I missed it during my initial review).
yup, the FPU registers handling has also been reworked a bit to accomodate XSAVE migration.
Quoted 4 lines
- the PDPTRs. That is probably okay (because you never set nor read
env->pdptrs_valid) but please check if Hyper-V supports reading them.I'll try to find that out. Haven't stumbled over this so far, but I understand it's relevant for 32bit PAE guest, which we probably haven't tested thoroughly yet.
Quoted 14 lines
- KVM also has support for migrating in the middle of an exception being delivered (for example if an EPT violation happens due to a write to the stack); this is handled with fields such as these:
VMSTATE_UINT8(env.exception_pending, X86CPU), VMSTATE_UINT8(env.exception_injected, X86CPU), VMSTATE_UINT8(env.exception_has_payload, X86CPU), VMSTATE_UINT64(env.exception_payload, X86CPU), VMSTATE_INT32(env.exception_nr, X86CPU), VMSTATE_INT32(env.interrupt_injected, X86CPU), VMSTATE_UINT8(env.soft_interrupt, X86CPU), VMSTATE_UINT8(env.nmi_injected, X86CPU), VMSTATE_UINT8(env.nmi_pending, X86CPU),Quoted 3 lines
please check how Hyper-V handles this situation so that it can be implemented in QEMU as well.
Those are covered in MSHV's "vCPU Events", I think they map quite cleanly to the QEMU representation:
https://docs.rs/mshv-bindings/0.6.7/src/mshv_bindings/x86_64/regs.rs.html#404
thanks,
magnus