Intel Processor Trace virtulization enabling
Subsystems (4)
v2 cover Intel Processor Trace virtulization enabling
From: Chao Peng <[email protected]>
Hi All,
Here is a patch-series which adding Processor Trace enabling in KVM guest. You can get It's software developer manuals from:
https://software.intel.com/sites/default/files/managed/c5/15/architecture-instruction-set-extensions-programming-reference.pdf
In Chapter 5 INTEL PROCESSOR TRACE: VMX IMPROVEMENTS.
Introduction:
Intel Processor Trace (Intel PT) is an extension of Intel Architecture that captures information about software execution using dedicated hardware facilities that cause only minimal performance perturbation to the software being traced. Details on the Intel PT infrastructure and trace capabilities can be found in the Intel 64 and IA-32 Architectures Software Developer’s Manual, Volume 3C.
The suite of architecture changes serve to simplify the process of virtualizing Intel PT for use by a guest software. There are two primary elements to this new architecture support for VMX support improvements made for Intel PT.
1. Addition of a new guest IA32_RTIT_CTL value field to the VMCS.
— This serves to speed and simplify the process of disabling trace on VM exit, and restoring it on VM entry.
2. Enabling use of EPT to redirect PT output.
— This enables the VMM to elect to virtualize the PT output buffer using EPT. In this mode, the CPU will treat PT output addresses as Guest Physical Addresses (GPAs) and translate them using EPT. This means that Intel PT output reads (of the ToPA table) and writes (of trace output) can cause EPT violations, and other output events.
Processor Trace virtualization can be work in one of 3 possible modes by set new option "pt_mode". Default value is host guest mode.
a. system-wide: trace both host/guest and output to host buffer;
b. host-only: only trace host and output to host buffer;
c. host-guest: trace host/guest simultaneous and output to their respective buffer.
Quoted 1 line
From v1:
- remove guest-only mode because guest-only mode can be covered by host-guest mode;
- always set "use GPA for processor tracing" in secondary execution control if it can be;
- trap RTIT_CTL read/write. Forbid write this msr when VMXON in L1 hypervisor.
Chao Peng (8):
perf/x86/intel/pt: Move Intel-PT MSR bit definitions to a public
header
perf/x86/intel/pt: change pt_cap_get() to a public function
KVM: x86: add Intel processor trace virtualization mode
KVM: x86: add Intel processor trace cpuid emulataion
KVM: x86: add Intel processor trace context for each vcpu
KVM: x86: Implement Intel processor trace context switch
KVM: x86: add Intel PT msr RTIT_CTL read/write
KVM: x86: Disable intercept for Intel processor trace MSRs
arch/x86/events/intel/pt.c | 3 +- arch/x86/events/intel/pt.h | 55 ----------- arch/x86/include/asm/intel_pt.h | 26 ++++++ arch/x86/include/asm/kvm_host.h | 1 + arch/x86/include/asm/msr-index.h | 35 +++++++ arch/x86/include/asm/vmx.h | 8 ++ arch/x86/kvm/cpuid.c | 23 ++++- arch/x86/kvm/svm.c | 6 ++ arch/x86/kvm/vmx.c | 195 ++++++++++++++++++++++++++++++++++++++- 9 files changed, 290 insertions(+), 62 deletions(-)
Signature · 1 line
1.8.3.1
PATCH v2 1/8 perf/x86/intel/pt: Move Intel-PT MSR bit definitions to a public header
From: Chao Peng <[email protected]>
Intel processor trace virtualization enabling in guest need to use these MSR bits, so move then to public header msr-index.h.
--- arch/x86/events/intel/pt.h | 37 ------------------------------------- arch/x86/include/asm/msr-index.h | 33 +++++++++++++++++++++++++++++++++ 2 files changed, 33 insertions(+), 37 deletions(-)
arch/x86/events/intel/pt.h +0 -37
@@ -20,43 +20,6 @@#define __INTEL_PT_H__/* -* PT MSR bit definitions -*/ -#define RTIT_CTL_TRACEEN BIT(0) -#define RTIT_CTL_CYCLEACC BIT(1) -#define RTIT_CTL_OS BIT(2) -#define RTIT_CTL_USR BIT(3) -#define RTIT_CTL_PWR_EVT_EN BIT(4) -#define RTIT_CTL_FUP_ON_PTW BIT(5) -#define RTIT_CTL_CR3EN BIT(7) -#define RTIT_CTL_TOPA BIT(8) -#define RTIT_CTL_MTC_EN BIT(9) -#define RTIT_CTL_TSC_EN BIT(10) -#define RTIT_CTL_DISRETC BIT(11) -#define RTIT_CTL_PTW_EN BIT(12) -#define RTIT_CTL_BRANCH_EN BIT(13) -#define RTIT_CTL_MTC_RANGE_OFFSET 14 -#define RTIT_CTL_MTC_RANGE (0x0full << RTIT_CTL_MTC_RANGE_OFFSET) -#define RTIT_CTL_CYC_THRESH_OFFSET 19 -#define RTIT_CTL_CYC_THRESH (0x0full << RTIT_CTL_CYC_THRESH_OFFSET) -#define RTIT_CTL_PSB_FREQ_OFFSET 24 -#define RTIT_CTL_PSB_FREQ (0x0full << RTIT_CTL_PSB_FREQ_OFFSET) -#define RTIT_CTL_ADDR0_OFFSET 32 -#define RTIT_CTL_ADDR0 (0x0full << RTIT_CTL_ADDR0_OFFSET) -#define RTIT_CTL_ADDR1_OFFSET 36 -#define RTIT_CTL_ADDR1 (0x0full << RTIT_CTL_ADDR1_OFFSET) -#define RTIT_CTL_ADDR2_OFFSET 40 -#define RTIT_CTL_ADDR2 (0x0full << RTIT_CTL_ADDR2_OFFSET) -#define RTIT_CTL_ADDR3_OFFSET 44 -#define RTIT_CTL_ADDR3 (0x0full << RTIT_CTL_ADDR3_OFFSET) -#define RTIT_STATUS_FILTEREN BIT(0) -#define RTIT_STATUS_CONTEXTEN BIT(1) -#define RTIT_STATUS_TRIGGEREN BIT(2) -#define RTIT_STATUS_BUFFOVF BIT(3) -#define RTIT_STATUS_ERROR BIT(4) -#define RTIT_STATUS_STOPPED BIT(5) - -/* * Single-entry ToPA: when this close to region boundary, switch * buffers to avoid losing data. */
arch/x86/include/asm/msr-index.h +33 -0
@@ -93,7 +93,40 @@#define MSR_PEBS_LD_LAT_THRESHOLD 0x000003f6#define MSR_IA32_RTIT_CTL 0x00000570 +#define RTIT_CTL_TRACEEN BIT(0) +#define RTIT_CTL_CYCLEACC BIT(1) +#define RTIT_CTL_OS BIT(2) +#define RTIT_CTL_USR BIT(3) +#define RTIT_CTL_PWR_EVT_EN BIT(4) +#define RTIT_CTL_FUP_ON_PTW BIT(5) +#define RTIT_CTL_CR3EN BIT(7) +#define RTIT_CTL_TOPA BIT(8) +#define RTIT_CTL_MTC_EN BIT(9) +#define RTIT_CTL_TSC_EN BIT(10) +#define RTIT_CTL_DISRETC BIT(11) +#define RTIT_CTL_PTW_EN BIT(12) +#define RTIT_CTL_BRANCH_EN BIT(13) +#define RTIT_CTL_MTC_RANGE_OFFSET 14 +#define RTIT_CTL_MTC_RANGE (0x0full << RTIT_CTL_MTC_RANGE_OFFSET) +#define RTIT_CTL_CYC_THRESH_OFFSET 19 +#define RTIT_CTL_CYC_THRESH (0x0full << RTIT_CTL_CYC_THRESH_OFFSET) +#define RTIT_CTL_PSB_FREQ_OFFSET 24 +#define RTIT_CTL_PSB_FREQ (0x0full << RTIT_CTL_PSB_FREQ_OFFSET) +#define RTIT_CTL_ADDR0_OFFSET 32 +#define RTIT_CTL_ADDR0 (0x0full << RTIT_CTL_ADDR0_OFFSET) +#define RTIT_CTL_ADDR1_OFFSET 36 +#define RTIT_CTL_ADDR1 (0x0full << RTIT_CTL_ADDR1_OFFSET) +#define RTIT_CTL_ADDR2_OFFSET 40 +#define RTIT_CTL_ADDR2 (0x0full << RTIT_CTL_ADDR2_OFFSET) +#define RTIT_CTL_ADDR3_OFFSET 44 +#define RTIT_CTL_ADDR3 (0x0full << RTIT_CTL_ADDR3_OFFSET)#define MSR_IA32_RTIT_STATUS 0x00000571 +#define RTIT_STATUS_FILTEREN BIT(0) +#define RTIT_STATUS_CONTEXTEN BIT(1) +#define RTIT_STATUS_TRIGGEREN BIT(2) +#define RTIT_STATUS_BUFFOVF BIT(3) +#define RTIT_STATUS_ERROR BIT(4) +#define RTIT_STATUS_STOPPED BIT(5)#define MSR_IA32_RTIT_ADDR0_A 0x00000580#define MSR_IA32_RTIT_ADDR0_B 0x00000581#define MSR_IA32_RTIT_ADDR1_A 0x00000582
Signature · 1 line
1.8.3.1
PATCH v2 2/8 perf/x86/intel/pt: change pt_cap_get() to a public function
In reply to Luwei KangFrom: Chao Peng <[email protected]>
Change pt_cap_get() to a public function so that KVM can access it.
--- arch/x86/events/intel/pt.c | 3 ++- arch/x86/events/intel/pt.h | 18 ------------------ arch/x86/include/asm/intel_pt.h | 20 ++++++++++++++++++++ 3 files changed, 22 insertions(+), 19 deletions(-)
arch/x86/events/intel/pt.c +2 -1
@@ -75,7 +75,7 @@ PT_CAP(psb_periods, 1, CPUID_EBX, 0xffff0000), }; -static u32 pt_cap_get (enum pt_capabilities cap) +u32 pt_cap_get (enum pt_capabilities cap) {struct pt_cap_desc *cd = &pt_caps[cap];u32 c = pt_pmu.caps[cd->leaf *PT_CPUID_REGS_NUM + cd->reg];
@@ -83,6 +83,7 @@ static u32 pt_cap_get(enum pt_capabilities cap)return (c & cd->mask) >> shift; } +EXPORT_SYMBOL_GPL (pt_cap_get); static ssize_t pt_cap_show(struct device *cdev, struct device_attribute *attr,
arch/x86/events/intel/pt.h +0 -18
@@ -51,24 +51,6 @@ struct topa_entry {/* TSC to Core Crystal Clock Ratio */ #define CPUID_TSC_LEAF 0x15 -enum pt_capabilities { - PT_CAP_max_subleaf =0 , - PT_CAP_cr3_filtering, - PT_CAP_psb_cyc, - PT_CAP_ip_filtering, - PT_CAP_mtc, - PT_CAP_ptwrite, - PT_CAP_power_event_trace, - PT_CAP_topa_output, - PT_CAP_topa_multiple_entries, - PT_CAP_single_range_output, - PT_CAP_payloads_lip, - PT_CAP_num_address_ranges, - PT_CAP_mtc_periods, - PT_CAP_cycle_thresholds, - PT_CAP_psb_periods, -}; -struct pt_pmu {struct pmu pmu;u32 caps[PT_CPUID_REGS_NUM *PT_CPUID_LEAVES ];
arch/x86/include/asm/intel_pt.h +20 -0
@@ -1,10 +1,30 @@#ifndef _ASM_X86_INTEL_PT_H#define _ASM_X86_INTEL_PT_H +enum pt_capabilities { + PT_CAP_max_subleaf =0 , + PT_CAP_cr3_filtering, + PT_CAP_psb_cyc, + PT_CAP_ip_filtering, + PT_CAP_mtc, + PT_CAP_ptwrite, + PT_CAP_power_event_trace, + PT_CAP_topa_output, + PT_CAP_topa_multiple_entries, + PT_CAP_single_range_output, + PT_CAP_payloads_lip, + PT_CAP_num_address_ranges, + PT_CAP_mtc_periods, + PT_CAP_cycle_thresholds, + PT_CAP_psb_periods, +}; +#if defined(CONFIG_PERF_EVENTS ) && defined(CONFIG_CPU_SUP_INTEL )void cpu_emergency_stop_pt (void ); +extern u32 pt_cap_get (enum pt_capabilities cap);#else static inline void cpu_emergency_stop_pt (void ) {} +static inline u32 pt_cap_get (enum pt_capabilities cap) {return 0 ; }#endif #endif /* _ASM_X86_INTEL_PT_H */
Signature · 1 line
1.8.3.1
PATCH v2 3/8 KVM: x86: add Intel processor trace virtualization mode
In reply to Luwei KangFrom: Chao Peng <[email protected]>
PT virtualization can be work in one of 3 possible modes:
a. system-wide: trace both host/guest and output to host buffer; b. host-only: only trace host and output to host buffer; c. host-guest: trace host/guest simultaneous and output to their
respective buffer.
--- arch/x86/include/asm/intel_pt.h | 6 ++++ arch/x86/include/asm/msr-index.h | 1 + arch/x86/include/asm/vmx.h | 6 ++++ arch/x86/kvm/vmx.c | 73 +++++++++++++++++++++++++++++++++++++--- 4 files changed, 82 insertions(+), 4 deletions(-)
arch/x86/include/asm/intel_pt.h +6 -0
@@ -1,6 +1,12 @@#ifndef _ASM_X86_INTEL_PT_H#define _ASM_X86_INTEL_PT_H +enum pt_mode { +PT_MODE_SYSTEM =0 , +PT_MODE_HOST , +PT_MODE_HOST_GUEST , +}; +enum pt_capabilities { PT_CAP_max_subleaf =0 , PT_CAP_cr3_filtering,
arch/x86/include/asm/msr-index.h +1 -0
@@ -751,6 +751,7 @@#define VMX_BASIC_INOUT 0x0040000000000000LLU/* MSR_IA32_VMX_MISC bits */ +#define MSR_IA32_VMX_MISC_INTEL_PT (1ULL << 14)#define MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS (1ULL << 29)#define MSR_IA32_VMX_MISC_PREEMPTION_TIMER_SCALE 0x1F/* AMD-V MSRs */
arch/x86/include/asm/vmx.h +6 -0
@@ -76,7 +76,9 @@#define SECONDARY_EXEC_SHADOW_VMCS 0x00004000#define SECONDARY_EXEC_RDSEED 0x00010000#define SECONDARY_EXEC_ENABLE_PML 0x00020000 +#define SECONDARY_EXEC_PT_CONCEAL_VMX 0x00080000#define SECONDARY_EXEC_XSAVES 0x00100000 +#define SECONDARY_EXEC_PT_USE_GPA 0x01000000#define SECONDARY_EXEC_TSC_SCALING 0x02000000#define PIN_BASED_EXT_INTR_MASK 0x00000001
@@ -97,6 +99,8 @@#define VM_EXIT_LOAD_IA32_EFER 0x00200000#define VM_EXIT_SAVE_VMX_PREEMPTION_TIMER 0x00400000#define VM_EXIT_CLEAR_BNDCFGS 0x00800000 +#define VM_EXIT_PT_SUPPRESS_PIP 0x01000000 +#define VM_EXIT_CLEAR_IA32_RTIT_CTL 0x02000000#define VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR 0x00036dff
@@ -108,6 +112,8 @@#define VM_ENTRY_LOAD_IA32_PAT 0x00004000#define VM_ENTRY_LOAD_IA32_EFER 0x00008000#define VM_ENTRY_LOAD_BNDCFGS 0x00010000 +#define VM_ENTRY_PT_SUPPRESS_PIP 0x00020000 +#define VM_ENTRY_LOAD_IA32_RTIT_CTL 0x00040000#define VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR 0x000011ff
arch/x86/kvm/vmx.c +69 -4
145 lines not on this page · raw message
Signature · 1 line
1.8.3.1
Paolo Bonzini
maintainerRe: [patch v2 3/8] KVM: x86: add Intel processor trace virtualization mode
arch/x86/include/asm/vmx.h line 116
Quoted 11 lines · Luwei Kang
arch/x86/include/asm/vmx.h +4
-0+#define VM_EXIT_PT_SUPPRESS_PIP 0x01000000 +#define VM_EXIT_CLEAR_IA32_RTIT_CTL 0x02000000#define VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR 0x00036dff@@ -108,6 +112,8 @@#define VM_ENTRY_LOAD_IA32_PAT 0x00004000#define VM_ENTRY_LOAD_IA32_EFER 0x00008000#define VM_ENTRY_LOAD_BNDCFGS 0x00010000 +#define VM_ENTRY_PT_SUPPRESS_PIP 0x00020000 +#define VM_ENTRY_LOAD_IA32_RTIT_CTL 0x00040000
Please use PT_CONCEAL instead of PT_SUPPRESS_PIP, to better match the SDM (for both vmexit and vmentry controls).
arch/x86/kvm/vmx.c static u32 vmx_exec_control(struct vcpu_vmx *vmx) · line 5318
Quoted 3 lines
arch/x86/kvm/vmx.c +3
-0+if (!enable_ept) + vmexit_control &= ~VM_EXIT_CLEAR_IA32_RTIT_CTL ; +
Why is this (and the similar bit-clear operation in vmx_vmentry_control) needed only for !enable_ept?
Shouldn't it be like
if (pt_mode == PT_MODE_SYSTEM) {
vmexit_control &= ~VM_EXIT_PT_SUPPRESS_PIP;
vmexit_control &= ~VM_EXIT_CLEAR_IA32_RTIT_CTL;
}and
if (pt_mode == PT_MODE_SYSTEM) {
vmentry_control &= ~VM_ENTRY_PT_SUPPRESS_PIP;
vmentry_control &= ~VM_ENTRY_LOAD_IA32_RTIT_CTL;
}Thanks,
Paolo
RE: [patch v2 3/8] KVM: x86: add Intel processor trace virtualization mode
Quoted 28 lines
arch/x86/include/asm/vmx.h line 116
Quoted 11 lines
arch/x86/include/asm/vmx.h +4
-0+#define VM_EXIT_PT_SUPPRESS_PIP 0x01000000 +#define VM_EXIT_CLEAR_IA32_RTIT_CTL 0x02000000#define VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR 0x00036dff@@ -108,6 +112,8 @@#define VM_ENTRY_LOAD_IA32_PAT 0x00004000#define VM_ENTRY_LOAD_IA32_EFER 0x00008000#define VM_ENTRY_LOAD_BNDCFGS 0x00010000 +#define VM_ENTRY_PT_SUPPRESS_PIP 0x00020000 +#define VM_ENTRY_LOAD_IA32_RTIT_CTL 0x00040000Please use
PT_CONCEALinstead ofPT_SUPPRESS_PIP, to better match the SDM (for both vmexit and vmentry controls).arch/x86/kvm/vmx.c static u32 vmx_exec_control(struct vcpu_vmx *vmx) · line 5318
Quoted 3 lines
arch/x86/kvm/vmx.c +3
-0+if (!enable_ept) + vmexit_control &= ~VM_EXIT_CLEAR_IA32_RTIT_CTL ; +Why is this (and the similar bit-clear operation in
vmx_vmentry_control) needed only for !enable_ept?Shouldn't it be like
if (pt_mode == PT_MODE_SYSTEM) { vmexit_control &= ~VM_EXIT_PT_SUPPRESS_PIP; vmexit_control &= ~VM_EXIT_CLEAR_IA32_RTIT_CTL; }
Quoted 7 lines
and
if (pt_mode == PT_MODE_SYSTEM) { vmentry_control &= ~VM_ENTRY_PT_SUPPRESS_PIP; vmentry_control &= ~VM_ENTRY_LOAD_IA32_RTIT_CTL; }
I think I have a misunderstand of " always set "use GPA for processor tracing" in secondary execution control if it can be ".
"use GPA for processor tracing" can't be set in SYSTEM mode even if hardware can set this bit. Because guest will still think this a GPA address and translate by EPT. In fact, RTIT_OUTPUT_BASE will always a HPA in SYSTEM mode. Will fix in next version.
Thanks,
Luwei Kang
Paolo Bonzini
maintainerRe: [patch v2 3/8] KVM: x86: add Intel processor trace virtualization mode
In reply to Luwei Kangarch/x86/kvm/vmx.c static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf) · line 3746
Quoted 4 lines · Luwei Kang
arch/x86/kvm/vmx.c +4
-0+if (!(_cpu_based_2nd_exec_control &SECONDARY_EXEC_PT_USE_GPA ) || + !(_vmexit_control &VM_EXIT_CLEAR_IA32_RTIT_CTL ) || + !(_vmentry_control &VM_ENTRY_LOAD_IA32_RTIT_CTL )) { + _cpu_based_2nd_exec_control &= ~SECONDARY_EXEC_PT_USE_GPA ;
Also, you are not checking anywhere if the SUPPRESS_PIP controls are available. This is probably the best place.
arch/x86/kvm/vmx.c static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf) · line 3748
Quoted 2 lines
arch/x86/kvm/vmx.c +2
-0+ _vmexit_control &= ~VM_EXIT_CLEAR_IA32_RTIT_CTL ; + _vmentry_control &= ~VM_ENTRY_LOAD_IA32_RTIT_CTL ;
These two are not needed; disabling SECONDARY_EXEC_PT_USE_GPA is enough. The tracing mode will revert to PT_SYSTEM, which does not use the load/clear RTIT_CTL controls.
Paolo
Quoted 2 lines
(quoted diff) +2
-0+ } +
RE: [patch v2 3/8] KVM: x86: add Intel processor trace virtualization mode
Quoted 6 lines
arch/x86/kvm/vmx.c static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf) · line 3746
Quoted 4 lines
arch/x86/kvm/vmx.c +4
-0+if (!(_cpu_based_2nd_exec_control &SECONDARY_EXEC_PT_USE_GPA ) || + !(_vmexit_control &VM_EXIT_CLEAR_IA32_RTIT_CTL ) || + !(_vmentry_control &VM_ENTRY_LOAD_IA32_RTIT_CTL )) { + _cpu_based_2nd_exec_control &= ~SECONDARY_EXEC_PT_USE_GPA ;Also, you are not checking anywhere if the
SUPPRESS_PIPcontrols are available. This is probably the best place.
SUPPRESS_PIP(should be "CONCEAL", will fix it.) is use for control of processor trace packet.
I think we should clear it when in SYSTEM mode (For example, PIPs are generated on VM exit, with NonRoot=0. On VM exit to SMM, VMCS packets are additionally generated). Why need check this here?
Quoted 7 lines
arch/x86/kvm/vmx.c static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf) · line 3748
Quoted 3 lines
arch/x86/kvm/vmx.c +2
-0+ _vmexit_control &= ~VM_EXIT_CLEAR_IA32_RTIT_CTL ; + _vmentry_control &= ~VM_ENTRY_LOAD_IA32_RTIT_CTL ;These two are not needed; disabling
SECONDARY_EXEC_PT_USE_GPAis enough.
The tracing mode will revert toPT_SYSTEM, which does not use the load/clearRTIT_CTLcontrols.
The status of *_RTIT_CTL should be same with SECONDARY_EXEC_PT_USE_GPA or would cause VM-entry failed. (architecture-instruction-set-extensions-programming-reference 5.2.3)
Paolo Bonzini
maintainerRe: [patch v2 3/8] KVM: x86: add Intel processor trace virtualization mode
Quoted 11 lines · Luwei
Quoted 6 lines
arch/x86/kvm/vmx.c static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf) · line 3746
Quoted 4 lines
arch/x86/kvm/vmx.c +4
-0+if (!(_cpu_based_2nd_exec_control &SECONDARY_EXEC_PT_USE_GPA ) || + !(_vmexit_control &VM_EXIT_CLEAR_IA32_RTIT_CTL ) || + !(_vmentry_control &VM_ENTRY_LOAD_IA32_RTIT_CTL )) { + _cpu_based_2nd_exec_control &= ~SECONDARY_EXEC_PT_USE_GPA ;Also, you are not checking anywhere if the
SUPPRESS_PIPcontrols are available. This is probably the best place.
SUPPRESS_PIP(should be "CONCEAL", will fix it.)is use for control of processor trace packet. I think we should clear it when in SYSTEM mode (For example, PIPs are
generated on VM exit, with NonRoot=0. On VM exit to SMM, VMCS packets
are additionally generated). Why need check this here?
I agree. But if SUPPRESS_PIP/CONCEAL is not available, you cannot enable HOST_GUEST mode I think?
Quoted 10 lines
Quoted 6 lines
arch/x86/kvm/vmx.c static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf) · line 3748
Quoted 3 lines
arch/x86/kvm/vmx.c +2
-0+ _vmexit_control &= ~VM_EXIT_CLEAR_IA32_RTIT_CTL ; + _vmentry_control &= ~VM_ENTRY_LOAD_IA32_RTIT_CTL ;These two are not needed; disabling
SECONDARY_EXEC_PT_USE_GPAis enough.
The tracing mode will revert toPT_SYSTEM, which does not use the load/clearRTIT_CTLcontrols.The status of
*_RTIT_CTLshould be same withSECONDARY_EXEC_PT_USE_GPAor would cause VM-entry failed. (architecture-instruction-set-extensions-programming-reference 5.2.3)
Indeed:
- disable "use GPA" here
- hardware_setup notices !cpu_has_vmx_pt_use_gpa() and switches tracing mode to system
- vmx_vcpu_setup (via vmx_vmentry_control and vmx_vmexit_control) turns off the IA32_RIT_CTL vmentry/vmexit controls (and also turns off SUPPRESS_PIP)
- everybody is happy
Paolo
PATCH v2 4/8 KVM: x86: add Intel processor trace cpuid emulataion
In reply to Luwei KangFrom: Chao Peng <[email protected]>
Expose Intel processor trace cpuid to guest.
--- arch/x86/include/asm/kvm_host.h | 1 + arch/x86/kvm/cpuid.c | 23 +++++++++++++++++++++-- arch/x86/kvm/svm.c | 6 ++++++ arch/x86/kvm/vmx.c | 6 ++++++ 4 files changed, 34 insertions(+), 2 deletions(-)
arch/x86/include/asm/kvm_host.h +1 -0
@@ -1004,6 +1004,7 @@ struct kvm_x86_ops {void (*handle_external_intr)(struct kvm_vcpu *vcpu);bool (*mpx_supported)(void );bool (*xsaves_supported)(void ); +bool (*pt_supported)(void );int (*check_nested_events)(struct kvm_vcpu *vcpu,bool external_intr);
arch/x86/kvm/cpuid.c +21 -2
@@ -70,6 +70,7 @@ u64 kvm_supported_xcr0(void)/* These are scattered features in cpufeatures.h. */ #define KVM_CPUID_BIT_AVX512_4VNNIW 2#define KVM_CPUID_BIT_AVX512_4FMAPS 3 +#define KVM_CPUID_BIT_INTEL_PT 25#define KF (x) bit(KVM_CPUID_BIT_##x)int kvm_update_cpuid (struct kvm_vcpu *vcpu)
@@ -327,6 +328,7 @@ static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,unsigned f_invpcid = kvm_x86_ops->invpcid_supported () ?F (INVPCID ) :0 ;unsigned f_mpx =kvm_mpx_supported () ?F (MPX ) :0 ;unsigned f_xsaves = kvm_x86_ops->xsaves_supported () ?F (XSAVES ) :0 ; +unsigned f_intel_pt = kvm_x86_ops->pt_supported () ?KF (INTEL_PT ) :0 ;/* cpuid 1.edx */ const u32 kvm_cpuid_1_edx_x86_features =
@@ -379,7 +381,7 @@ static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,F (BMI2 ) |F (ERMS ) | f_invpcid |F (RTM ) | f_mpx |F (RDSEED ) |F (ADX ) |F (SMAP ) |F (AVX512IFMA ) |F (AVX512F ) |F (AVX512PF ) |F (AVX512ER ) |F (AVX512CD ) |F (CLFLUSHOPT ) |F (CLWB ) |F (AVX512DQ ) | -F (SHA_NI ) |F (AVX512BW ) |F (AVX512VL ); +F (SHA_NI ) |F (AVX512BW ) |F (AVX512VL ) | f_intel_pt;/* cpuid 0xD.1.eax */ const u32 kvm_cpuid_D_1_eax_x86_features =
@@ -407,7 +409,7 @@ static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,switch (function) {case 0 : - entry->eax =min (entry->eax, (u32 )0xd ); + entry->eax =min (entry->eax, (u32)(f_intel_pt ?0x14 :0xd ));break ;case 1 : entry->edx &= kvm_cpuid_1_edx_x86_features;
@@ -578,6 +580,23 @@ static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function, }break ; } +/* Intel PT */ +case 0x14 : { +int t, times = entry->eax &0xffffffff ; + +if (!f_intel_pt) +break ; + + entry->flags |=KVM_CPUID_FLAG_SIGNIFCANT_INDEX ; +for (t =1 ; t <= times; ++t) { +if (*nent >= maxnent) + goto out; +do_cpuid_1_ent (&entry[t], function, t); + entry[t].flags |=KVM_CPUID_FLAG_SIGNIFCANT_INDEX ; + ++*nent; + } +break ; + }case KVM_CPUID_SIGNATURE : {static const char signature[12 ] ="KVMKVMKVM\0\0" ;const u32 *sigptr = (const u32 *)signature;
arch/x86/kvm/svm.c +6 -0
@@ -5174,6 +5174,11 @@ static bool svm_xsaves_supported(void)return false; } +static bool svm_pt_supported (void ) +{ +return false; +} +static bool svm_has_wbinvd_exit (void ) {return true;
@@ -5485,6 +5490,7 @@ static void svm_setup_mce(struct kvm_vcpu *vcpu) .invpcid_supported = svm_invpcid_supported, .mpx_supported = svm_mpx_supported, .xsaves_supported = svm_xsaves_supported, + .pt_supported = svm_pt_supported, .set_supported_cpuid = svm_set_supported_cpuid,
arch/x86/kvm/vmx.c +6 -0
@@ -9163,6 +9163,11 @@ static bool vmx_xsaves_supported(void)SECONDARY_EXEC_XSAVES ; } +static bool vmx_pt_supported (void ) +{ +return (pt_mode ==PT_MODE_HOST_GUEST ); +} +static void vmx_recover_nmi_blocking (struct vcpu_vmx *vmx) {u32 exit_intr_info;
@@ -12107,6 +12112,7 @@ static void vmx_setup_mce(struct kvm_vcpu *vcpu) .handle_external_intr = vmx_handle_external_intr, .mpx_supported = vmx_mpx_supported, .xsaves_supported = vmx_xsaves_supported, + .pt_supported = vmx_pt_supported, .check_nested_events = vmx_check_nested_events,
Signature · 1 line
1.8.3.1
Paolo Bonzini
maintainerRe: [patch v2 4/8] KVM: x86: add Intel processor trace cpuid emulataion
arch/x86/kvm/cpuid.c u64 kvm_supported_xcr0(void) · line 73
Quoted 9 lines · Luwei Kang
arch/x86/kvm/cpuid.c +1
-0@@ -70,6 +70,7 @@ u64 kvm_supported_xcr0(void)/* These are scattered features in cpufeatures.h. */ #define KVM_CPUID_BIT_AVX512_4VNNIW 2#define KVM_CPUID_BIT_AVX512_4FMAPS 3 +#define KVM_CPUID_BIT_INTEL_PT 25
This is not necessary, because there is no need to place processor tracing in scattered features. Can you replace this hunk, and the KF usage below, with the following patch?
Thanks,
Paolo
------------ 8< -------------
From: Paolo Bonzini <[email protected]>
Subject: [PATCH] x86: cpufeature: move processor tracing out of scattered features
Processor tracing is already enumerated in word 9 (CPUID[7,0].EBX), so do not duplicate it in the scattered features word.
--- arch/x86/include/asm/cpufeatures.h | 3 ++- arch/x86/kernel/cpu/scattered.c | 1 - 2 files changed, 2 insertions(+), 2 deletions(-)
arch/x86/include/asm/cpufeatures.h +2 -1
@@ -199,7 +199,7 @@#define X86_FEATURE_SME ( 7*32+10)/* AMD Secure Memory Encryption */ #define X86_FEATURE_INTEL_PPIN ( 7*32+14)/* Intel Processor Inventory Number */ -#define X86_FEATURE_INTEL_PT ( 7*32+15)/* Intel Processor Trace */ +#define X86_FEATURE_AVX512_4VNNIW (7*32+16)/* AVX-512 Neural Network Instructions */ #define X86_FEATURE_AVX512_4FMAPS (7*32+17)/* AVX-512 Multiply Accumulation Single precision */
@@ -238,6 +238,7 @@#define X86_FEATURE_AVX512IFMA ( 9*32+21)/* AVX-512 Integer Fused Multiply-Add instructions */ #define X86_FEATURE_CLFLUSHOPT ( 9*32+23)/* CLFLUSHOPT instruction */ #define X86_FEATURE_CLWB ( 9*32+24)/* CLWB instruction */ +#define X86_FEATURE_INTEL_PT ( 9*32+25)/* Intel Processor Trace */ #define X86_FEATURE_AVX512PF ( 9*32+26)/* AVX-512 Prefetch */ #define X86_FEATURE_AVX512ER ( 9*32+27)/* AVX-512 Exponential and Reciprocal */ #define X86_FEATURE_AVX512CD ( 9*32+28)/* AVX-512 Conflict Detection */
arch/x86/kernel/cpu/scattered.c +0 -1
@@ -21,7 +21,6 @@ struct cpuid_bit {static const struct cpuid_bit cpuid_bits[] = { {X86_FEATURE_APERFMPERF ,CPUID_ECX ,0 ,0x00000006 ,0 }, {X86_FEATURE_EPB ,CPUID_ECX ,3 ,0x00000006 ,0 }, - {X86_FEATURE_INTEL_PT ,CPUID_EBX ,25 ,0x00000007 ,0 }, {X86_FEATURE_AVX512_4VNNIW ,CPUID_EDX ,2 ,0x00000007 ,0 }, {X86_FEATURE_AVX512_4FMAPS ,CPUID_EDX ,3 ,0x00000007 ,0 }, {X86_FEATURE_CAT_L3 ,CPUID_EBX ,1 ,0x00000010 ,0 },
Quoted 11 lines
arch/x86/kvm/cpuid.c +1
-0#define KF (x) bit(KVM_CPUID_BIT_##x)int kvm_update_cpuid (struct kvm_vcpu *vcpu)@@ -327,6 +328,7 @@ static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,unsigned f_invpcid = kvm_x86_ops->invpcid_supported () ?F (INVPCID ) :0 ;unsigned f_mpx =kvm_mpx_supported () ?F (MPX ) :0 ;unsigned f_xsaves = kvm_x86_ops->xsaves_supported () ?F (XSAVES ) :0 ; +unsigned f_intel_pt = kvm_x86_ops->pt_supported () ?KF (INTEL_PT ) :0 ;/* cpuid 1.edx */ const u32 kvm_cpuid_1_edx_x86_features =
RE: [patch v2 4/8] KVM: x86: add Intel processor trace cpuid emulataion
Quoted 12 lines
arch/x86/kvm/cpuid.c u64 kvm_supported_xcr0(void) · line 73
Quoted 9 lines
arch/x86/kvm/cpuid.c index renamed from arch/x86/kvm/cpuid.c +0
-00099e10..ef19a11 100644
arch/x86/kvm/cpuid.c +1
-0@@ -70,6 +70,7 @@ u64 kvm_supported_xcr0(void)/* These are scattered features in cpufeatures.h. */ #define KVM_CPUID_BIT_AVX512_4VNNIW 2#define KVM_CPUID_BIT_AVX512_4FMAPS 3 +#define KVM_CPUID_BIT_INTEL_PT 25This is not necessary, because there is no need to place processor tracing in scattered features. Can you replace this hunk, and the KF usage below, with the following patch?
Yes, this looks good to me. will fix in next version.
Thanks,
Luwei Kang
Quoted 57 lines
------------ 8< -------------
From: Paolo Bonzini <[email protected]>
Subject: [PATCH] x86: cpufeature: move processor tracing out of scattered featuresProcessor tracing is already enumerated in word 9 (CPUID[7,0].EBX), so do not duplicate it in the scattered features word.
Signed-off-by Paolo Bonzini <[email protected]>--- arch/x86/include/asm/cpufeatures.h | 3 ++- arch/x86/kernel/cpu/scattered.c | 1 - 2 files changed, 2 insertions(+), 2 deletions(-)arch/x86/include/asm/cpufeatures.h +1
-1@@ -199,7 +199,7 @@#define X86_FEATURE_SME ( 7*32+10)/* AMD Secure Memory Encryption */ #define X86_FEATURE_INTEL_PPIN ( 7*32+14)/* Intel Processor Inventory Number */ -#define X86_FEATURE_INTEL_PT ( 7*32+15)/* Intel Processor Trace */ + #define X86_FEATURE_AVX512_4VNNIW (7*32+16) /* AVX-512 Neural Network Instructions */ #define
X86_FEATURE_AVX512_4FMAPS(7*32+17) /* AVX-512 Multiply Accumulation Single precision */arch/x86/include/asm/cpufeatures.h +1
-0@@ -238,6 +238,7 @@#define X86_FEATURE_AVX512IFMA ( 9*32+21)/* AVX-512 Integer Fused Multiply-Add instructions */ #define X86_FEATURE_CLFLUSHOPT ( 9*32+23)/* CLFLUSHOPT instruction */ #define X86_FEATURE_CLWB ( 9*32+24)/* CLWB instruction */ +#define X86_FEATURE_INTEL_PT ( 9*32+25)/* Intel Processor Trace */ #define X86_FEATURE_AVX512PF ( 9*32+26)/* AVX-512 Prefetch */ #define X86_FEATURE_AVX512ER ( 9*32+27)/* AVX-512 Exponential and Reciprocal */ #define X86_FEATURE_AVX512CD ( 9*32+28)/* AVX-512 Conflict Detection */ arch/x86/kernel/cpu/scattered.c +0
-1@@ -21,7 +21,6 @@ struct cpuid_bit {static const struct cpuid_bit cpuid_bits[] = { {X86_FEATURE_APERFMPERF ,CPUID_ECX ,0 ,0x00000006 ,0 }, {X86_FEATURE_EPB ,CPUID_ECX ,3 ,0x00000006 ,0 }, - {X86_FEATURE_INTEL_PT ,CPUID_EBX ,25 ,0x00000007 ,0 }, {X86_FEATURE_AVX512_4VNNIW ,CPUID_EDX ,2 ,0x00000007 ,0 }, {X86_FEATURE_AVX512_4FMAPS ,CPUID_EDX ,3 ,0x00000007 ,0 }, {X86_FEATURE_CAT_L3 ,CPUID_EBX ,1 ,0x00000010 ,0 },Quoted 13 lines
arch/x86/kvm/cpuid.c +0
-0#define KF (x) bit(KVM_CPUID_BIT_##x) int kvm_update_cpuid(struct kvm_vcpu *vcpu) @@ -327,6 +328,7 @@static inline int __do_cpuid_ent(struct
kvm_cpuid_entry2*entry, u32 function,arch/x86/kvm/cpuid.c +2
-0unsigned f_invpcid = kvm_x86_ops->invpcid_supported () ?F (INVPCID ) :0 ;unsigned f_mpx =kvm_mpx_supported () ?F (MPX ) :0 ;unsigned f_xsaves = kvm_x86_ops->xsaves_supported () ?F (XSAVES ) :0 ; +unsigned f_intel_pt = kvm_x86_ops->pt_supported () ?KF (INTEL_PT ) : +0 ;/* cpuid 1.edx */ const u32 kvm_cpuid_1_edx_x86_features =
PATCH v2 5/8 KVM: x86: add Intel processor trace context for each vcpu
In reply to Luwei KangFrom: Chao Peng <[email protected]>
Add a date structure to save Intel processor trace context. It mainly include all the MSR of Intel processor trace.
--- arch/x86/include/asm/msr-index.h | 1 + arch/x86/include/asm/vmx.h | 2 ++ arch/x86/kvm/vmx.c | 17 +++++++++++++++++ 3 files changed, 20 insertions(+)
arch/x86/include/asm/msr-index.h +1 -0
@@ -135,6 +135,7 @@#define MSR_IA32_RTIT_ADDR2_B 0x00000585#define MSR_IA32_RTIT_ADDR3_A 0x00000586#define MSR_IA32_RTIT_ADDR3_B 0x00000587 +#define MSR_IA32_RTIT_ADDR_COUNT 8#define MSR_IA32_RTIT_CR3_MATCH 0x00000572#define MSR_IA32_RTIT_OUTPUT_BASE 0x00000560#define MSR_IA32_RTIT_OUTPUT_MASK 0x00000561
arch/x86/include/asm/vmx.h +2 -0
@@ -240,6 +240,8 @@ enum vmcs_field {GUEST_PDPTR3_HIGH =0x00002811 ,GUEST_BNDCFGS =0x00002812 ,GUEST_BNDCFGS_HIGH =0x00002813 , +GUEST_IA32_RTIT_CTL =0x00002814 , +GUEST_IA32_RTIT_CTL_HIGH =0x00002815 ,HOST_IA32_PAT =0x00002c00 ,HOST_IA32_PAT_HIGH =0x00002c01 , HOST_IA32_EFER = 0x00002c02,
arch/x86/kvm/vmx.c +17 -0
@@ -566,6 +566,21 @@ static inline int pi_test_sn(struct pi_desc *pi_desc) (unsigned long *)&pi_desc->control); } +struct pt_ctx { +u64 ctl; +u64 status; +u64 output_base; +u64 output_mask; +u64 cr3_match; +u64 addrs[MSR_IA32_RTIT_ADDR_COUNT ]; +}; + +struct pt_desc { +unsigned int addr_num; +struct pt_ctx host; +struct pt_ctx guest; +}; +struct vcpu_vmx {struct kvm_vcpu vcpu;unsigned long host_rsp;
@@ -655,6 +670,8 @@ struct vcpu_vmx {*/ u64 msr_ia32_feature_control;u64 msr_ia32_feature_control_valid_bits; + +struct pt_desc pt_desc; };enum segment_cache_field {
Signature · 1 line
1.8.3.1
PATCH v2 6/8 KVM: x86: Implement Intel processor trace context switch
In reply to Luwei KangFrom: Chao Peng <[email protected]>
Load/Store Intel processor trace register in context switch. MSR IA32_RTIT_CTL is loaded/stored automatically from VMCS, other MSRs are loaded/stored manaully.
--- arch/x86/kvm/vmx.c | 75 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 75 insertions(+)
arch/x86/kvm/vmx.c +75 -0
@@ -2131,6 +2131,61 @@ static unsigned long segment_base(u16 selector) }#endif +static inline void pt_load_msr (struct pt_ctx *ctx,unsigned int addr_num) +{ +u32 i; + +wrmsrl (MSR_IA32_RTIT_STATUS , ctx->status); +wrmsrl (MSR_IA32_RTIT_OUTPUT_BASE , ctx->output_base); +wrmsrl (MSR_IA32_RTIT_OUTPUT_MASK , ctx->output_mask); +wrmsrl (MSR_IA32_RTIT_CR3_MATCH , ctx->cr3_match); +for (i =0 ; i < addr_num; i++) +wrmsrl (MSR_IA32_RTIT_ADDR0_A + i, ctx->addrs[i]); +} + +static inline void pt_save_msr (struct pt_ctx *ctx,unsigned int addr_num) +{ +u32 i; + +rdmsrl (MSR_IA32_RTIT_STATUS , ctx->status); +rdmsrl (MSR_IA32_RTIT_OUTPUT_BASE , ctx->output_base); +rdmsrl (MSR_IA32_RTIT_OUTPUT_MASK , ctx->output_mask); +rdmsrl (MSR_IA32_RTIT_CR3_MATCH , ctx->cr3_match); +for (i =0 ; i < addr_num; i++) +rdmsrl (MSR_IA32_RTIT_ADDR0_A + i, ctx->addrs[i]); +} + +static void pt_guest_enter (struct vcpu_vmx *vmx) +{ +u64 ctl; + +if (pt_mode ==PT_MODE_HOST || pt_mode ==PT_MODE_HOST_GUEST ) { +rdmsrl (MSR_IA32_RTIT_CTL , ctl); + vmx->pt_desc.host.ctl = ctl; +if (ctl &RTIT_CTL_TRACEEN ) { + ctl &= ~RTIT_CTL_TRACEEN ; +wrmsrl (MSR_IA32_RTIT_CTL , ctl); + } + } + +if (pt_mode ==PT_MODE_HOST_GUEST ) { +pt_save_msr (&vmx->pt_desc.host, vmx->pt_desc.addr_num); +pt_load_msr (&vmx->pt_desc.guest, vmx->pt_desc.addr_num); + } +} + +static void pt_guest_exit (struct vcpu_vmx *vmx) +{ +if (pt_mode ==PT_MODE_HOST_GUEST ) { +pt_save_msr (&vmx->pt_desc.guest, vmx->pt_desc.addr_num); +pt_load_msr (&vmx->pt_desc.host, vmx->pt_desc.addr_num); +wrmsrl (MSR_IA32_RTIT_CTL , vmx->pt_desc.host.ctl); + } + +if (pt_mode ==PT_MODE_HOST ) +wrmsrl (MSR_IA32_RTIT_CTL , vmx->pt_desc.host.ctl); +} +static void vmx_save_host_state (struct kvm_vcpu *vcpu) {struct vcpu_vmx *vmx =to_vmx (vcpu);
@@ -5618,6 +5673,13 @@ static int vmx_vcpu_setup(struct vcpu_vmx *vmx)vmcs_write16 (GUEST_PML_INDEX ,PML_ENTITY_NUM -1 ); } +if (pt_mode ==PT_MODE_HOST_GUEST ) { +u32 eax, ebx, ecx, edx; + +cpuid_count (0x14 ,1 , &eax, &ebx, &ecx, &edx); + vmx->pt_desc.addr_num = eax &0x7 ; + } +return 0 ; }
@@ -5712,6 +5774,15 @@ static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)update_exception_bitmap (vcpu);vpid_sync_context (vmx->vpid); + +if (pt_mode ==PT_MODE_HOST_GUEST ) { +memset (&vmx->pt_desc.host,0 ,sizeof (vmx->pt_desc.host)); +memset (&vmx->pt_desc.guest,0 ,sizeof (vmx->pt_desc.guest)); +/* Bit[6~0] are forced to 1, writes are ignored. */ + vmx->pt_desc.guest.output_mask =0x7F ; +/* Clear the status of IA32_RTIT_CTL in VMCS guest state. */ +vmcs_write32 (GUEST_IA32_RTIT_CTL ,0 ); + } }/*
@@ -9382,6 +9453,8 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu) vcpu->arch.pkru != vmx->host_pkru)__write_pkru (vcpu->arch.pkru); +pt_guest_enter (vmx); +atomic_switch_perf_msrs (vmx); debugctlmsr =get_debugctlmsr ();
@@ -9517,6 +9590,8 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu) | (1 << VCPU_EXREG_CR3)); vcpu->arch.regs_dirty =0 ; +pt_guest_exit (vmx); +/* * eager fpu is enabled if PKEY is supported and CR4 is switched * back on host, so it is safe to read guest PKRU from current
Signature · 1 line
1.8.3.1
Paolo Bonzini
maintainerRe: [patch v2 6/8] KVM: x86: Implement Intel processor trace context switch
arch/x86/kvm/vmx.c static unsigned long segment_base(u16 selector) · line 2168
Quoted 11 lines · Luwei Kang
arch/x86/kvm/vmx.c +11
-0+static void pt_guest_enter (struct vcpu_vmx *vmx) +{ +u64 ctl; + +if (pt_mode ==PT_MODE_HOST || pt_mode ==PT_MODE_HOST_GUEST ) { +rdmsrl (MSR_IA32_RTIT_CTL , ctl); + vmx->pt_desc.host.ctl = ctl; +if (ctl &RTIT_CTL_TRACEEN ) { + ctl &= ~RTIT_CTL_TRACEEN ; +wrmsrl (MSR_IA32_RTIT_CTL , ctl); + }
This "if" is only needed for PT_MODE_HOST_GUEST, I believe. PT_MODE_HOST can just use the "load RTIT_CTL" vmentry control to disable tracing.
Quoted 9 lines
arch/x86/kvm/vmx.c +9
-0+ } + +if (pt_mode ==PT_MODE_HOST_GUEST ) { +pt_save_msr (&vmx->pt_desc.host, vmx->pt_desc.addr_num); +pt_load_msr (&vmx->pt_desc.guest, vmx->pt_desc.addr_num); + } +} + +static void pt_guest_exit (struct vcpu_vmx *vmx)
arch/x86/kvm/vmx.c static unsigned long segment_base(u16 selector) · line 2187
Quoted 10 lines
arch/x86/kvm/vmx.c +10
-0+{ +if (pt_mode ==PT_MODE_HOST_GUEST ) { +pt_save_msr (&vmx->pt_desc.guest, vmx->pt_desc.addr_num); +pt_load_msr (&vmx->pt_desc.host, vmx->pt_desc.addr_num); +wrmsrl (MSR_IA32_RTIT_CTL , vmx->pt_desc.host.ctl); + } + +if (pt_mode ==PT_MODE_HOST ) +wrmsrl (MSR_IA32_RTIT_CTL , vmx->pt_desc.host.ctl); +}
Please use an
if (pt_mode == PT_MODE_HOST || pt_mode == PT_MODE_HOST_GUEST)
for the write to RTIT_CTL, so that pt_guest_exit mirrors pt_guest_entry.
Also, we don't actually need to write the MSR if RTIT_CTL_TRACEEN is false. With these changes, the cost of the "host-only" mode is acceptable, but for host-guest mode it is very expensive to read and write the MSRs on all vmentries and vmexits is very expensive.
It would be much better to avoid writing the guest state if the guest RTIT_CTL has TRACEEN=0. This would require keeping the intercepts until TRACEEN=1, but a lot of the work would be needed anyway---see my review of patch 7.
Thanks,
Paolo
PATCH v2 7/8 KVM: x86: add Intel PT msr RTIT_CTL read/write
In reply to Luwei KangFrom: Chao Peng <[email protected]>
L1 hypervisor can't get the capability of "TraceEn can be set in VMX operation"(IA32_VMX_MISC[14]=0) and any attempt to set IA32_RTIT_CTL.TraceEn in VMX operation using WRMSR will cause a general-protection exception if IA32_VMX_MISC[14] is 0. So we need to leave out write this msr in L1 VMX operation.
--- arch/x86/kvm/vmx.c | 11 +++++++++++ 1 file changed, 11 insertions(+)
arch/x86/kvm/vmx.c +11 -0
@@ -936,6 +936,7 @@ static void vmx_get_segment(struct kvm_vcpu *vcpu,static void vmx_set_nmi_mask (struct kvm_vcpu *vcpu,bool masked);static bool nested_vmx_is_page_fault_vmexit (struct vmcs12 *vmcs12,u16 error_code); +static bool vmx_pt_supported (void ); static DEFINE_PER_CPU(struct vmcs *, vmxarea); static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
@@ -3384,6 +3385,11 @@ static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)return 1 ; msr_info->data = vcpu->arch.ia32_xss;break ; +case MSR_IA32_RTIT_CTL : +if (!vmx_pt_supported ()) +return 1 ; + msr_info->data =vmcs_read64 (GUEST_IA32_RTIT_CTL ); +break ;case MSR_TSC_AUX :if (!msr_info->host_initiated && !guest_cpuid_has (vcpu,X86_FEATURE_RDTSCP ))
@@ -3508,6 +3514,11 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)else clear_atomic_switch_msr (vmx,MSR_IA32_XSS );break ; +case MSR_IA32_RTIT_CTL : +if (!vmx_pt_supported () ||to_vmx (vcpu)->nested.vmxon) +return 1 ; +vmcs_write64 (GUEST_IA32_RTIT_CTL , data); +break ;case MSR_TSC_AUX :if (!msr_info->host_initiated && !guest_cpuid_has (vcpu,X86_FEATURE_RDTSCP ))
Signature · 1 line
1.8.3.1
Paolo Bonzini
maintainerRe: [patch v2 7/8] KVM: x86: add Intel PT msr RTIT_CTL read/write
Quoted 40 lines · Luwei Kang
From: Chao Peng <[email protected]>
L1 hypervisor can't get the capability of "TraceEn can be set in VMX operation"(IA32_VMX_MISC[14]=0) and any attempt to set IA32_RTIT_CTL.TraceEn in VMX operation using WRMSR will cause a general-protection exception if IA32_VMX_MISC[14] is 0. So we need to leave out write this msr in L1 VMX operation.
Signed-off-by Chao Peng <[email protected]>Signed-off-by Luwei Kang <[email protected]>--- arch/x86/kvm/vmx.c | 11 +++++++++++ 1 file changed, 11 insertions(+)arch/x86/kvm/vmx.c +6
-0@@ -936,6 +936,7 @@ static void vmx_get_segment(struct kvm_vcpu *vcpu,static void vmx_set_nmi_mask (struct kvm_vcpu *vcpu,bool masked);static bool nested_vmx_is_page_fault_vmexit (struct vmcs12 *vmcs12,u16 error_code); +static bool vmx_pt_supported (void ); static DEFINE_PER_CPU(struct vmcs *, vmxarea); static DEFINE_PER_CPU(struct vmcs *, current_vmcs);@@ -3384,6 +3385,11 @@ static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)return 1 ; msr_info->data = vcpu->arch.ia32_xss;break ; +case MSR_IA32_RTIT_CTL : +if (!vmx_pt_supported ()) +return 1 ; + msr_info->data =vmcs_read64 (GUEST_IA32_RTIT_CTL ); +break ;case MSR_TSC_AUX :if (!msr_info->host_initiated && !guest_cpuid_has (vcpu,X86_FEATURE_RDTSCP ))
arch/x86/kvm/vmx.c static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) · line 3519
Quoted 6 lines
arch/x86/kvm/vmx.c +3
-0@@ -3508,6 +3514,11 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)else clear_atomic_switch_msr (vmx,MSR_IA32_XSS );break ; +case MSR_IA32_RTIT_CTL : +if (!vmx_pt_supported () ||to_vmx (vcpu)->nested.vmxon) +return 1 ;
VMXON must also clear TraceEn bit (see 23.4 in the SDM).
You also need to support R/W of all the other MSRs, in order to make them accessible to userspace, and add them in msrs_to_save and kvm_init_msr_list.
Regarding the save/restore of the state, I am now wondering if you could also use XSAVES and XRSTORS instead of multiple rdmsr/wrmsr in a loop. The cost is two MSR writes on vmenter (because the guest must run with IA32_XSS=0) and two on vmexit.
Thanks,
Paolo
Quoted 6 lines
arch/x86/kvm/vmx.c +2
-0+vmcs_write64 (GUEST_IA32_RTIT_CTL , data); +break ;case MSR_TSC_AUX :if (!msr_info->host_initiated && !guest_cpuid_has (vcpu,X86_FEATURE_RDTSCP ))
RE: [patch v2 7/8] KVM: x86: add Intel PT msr RTIT_CTL read/write
Quoted 15 lines
Quoted 15 lines
static
DEFINE_PER_CPU(struct vmcs *, vmxarea); staticDEFINE_PER_CPU(struct vmcs *, current_vmcs); @@ -3384,6 +3385,11 @@ static intvmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)arch/x86/kvm/vmx.c +5
-0return 1 ; msr_info->data = vcpu->arch.ia32_xss;break ; +case MSR_IA32_RTIT_CTL : +if (!vmx_pt_supported ()) +return 1 ; + msr_info->data =vmcs_read64 (GUEST_IA32_RTIT_CTL ); +break ;case MSR_TSC_AUX :if (!msr_info->host_initiated && !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP)) @@ -3508,6 +3514,11@@ static int
vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
Quoted 8 lines
arch/x86/kvm/vmx.c static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) · line 3519
Quoted 6 lines
arch/x86/kvm/vmx.c +3
-0else clear_atomic_switch_msr (vmx,MSR_IA32_XSS );break ; +case MSR_IA32_RTIT_CTL : +if (!vmx_pt_supported () ||to_vmx (vcpu)->nested.vmxon) +return 1 ;VMXON must also clear TraceEn bit (see 23.4 in the SDM).
Will clear TraceEn bit in handle_vmon() function.
Quoted 3 lines
You also need to support R/W of all the other MSRs, in order to make them accessible to userspace, and add them in
msrs_to_saveandkvm_init_msr_list.
Will add it in next version. This is use for live migration, is that right?
Quoted 3 lines
Regarding the save/restore of the state, I am now wondering if you could also use XSAVES and XRSTORS instead of multiple rdmsr/wrmsr in a loop. The cost is two MSR writes on vmenter (because the guest must run with IA32_XSS=0) and two on vmexit.
If use XSAVES and XRSTORS for context switch.
1. Before kvm_load_guest_fpu(vcpu), we need to save host RTIT_CTL, disable PT and restore the value of "vmx->pt_desc.guest.ctl" to GUEST_IA32_RTIT_CTL. Is that right?
2. After VM-exit (step out from kvm_x86_ops->run(vcpu)), we need to save the status of GUEST_IA32_RTIT_CTL . TRACEEN=0 and others MSRs are still in guest status. Where to enable PT if in host-guest mode. I think we should enable PT after vm-exit but it may cause #GP. " If XRSTORS would restore (or initialize) PT state and IA32_RTIT_CTL.TraceEn = 1, the instruction causes a general-protection exception. SDM 13.5.6". if enable after kvm_put_guest_fpu() I think it too late.)
Thanks,
Luwei Kang
Quoted 7 lines
Quoted 7 lines
arch/x86/kvm/vmx.c +2
-0+vmcs_write64 (GUEST_IA32_RTIT_CTL , data); +break ;case MSR_TSC_AUX :if (!msr_info->host_initiated && !guest_cpuid_has (vcpu,X86_FEATURE_RDTSCP ))
Paolo Bonzini
maintainerRe: [patch v2 7/8] KVM: x86: add Intel PT msr RTIT_CTL read/write
Quoted 1 line · Luwei
Will add it in next version. This is use for live migration, is that right?
Yes.
Quoted 2 lines
If use XSAVES and XRSTORS for context switch.
1. Beforekvm_load_guest_fpu(vcpu), we need to save hostRTIT_CTL, disable PT and restore the value of "vmx->pt_desc.guest.ctl" toGUEST_IA32_RTIT_CTL. Is that right?
The idea is to make the MSR get/set handlers operate on an XSAVES state separate from the guest FPU state. RTIT_CTL in the XSAVES state is special cased to always be zero.
Then you could do
if (!system mode) {
if (guest RTIT_CTL.TraceEn != 0) {
set PT in IA32_XSS
XSAVES the host PT state
// RTIT_CTL.TraceEn is now zero, and remains zero after XRSTORS
XRSTORS the guest PT state
clear PT in IA32_XSS
} else {
save host RTIT_CTL
}
// guest RTIT_CTL.TraceEn will be loaded by vmentry
}on vmentry, and
if (!system mode) {
// RTIT_CTL.TraceEn is zero here
if (guest RTIT_CTL.TraceEn != 0) {
set PT in IA32_XSS
// no need to XSAVES guest state, all MSR writes cause a vmexit
XRSTORS the host PT state
clear PT in IA32_XSS
} else if (host RTIT_CTL.TraceEn != 0) {
restore host RTIT_CTL
}
}on vmexit. It should still be cheaper than many rdmsr/wrmsr operations.
Paolo
Quoted 7 lines
2. After VM-exit (step out from
kvm_x86_ops->run(vcpu)), we need to save the status ofGUEST_IA32_RTIT_CTL. TRACEEN=0 and others MSRs are still in guest status. Where to enable PT if in host-guest mode. I think we should enable PT after vm-exit but it may cause #GP. " If XRSTORS would restore (or initialize) PT state and IA32_RTIT_CTL.TraceEn = 1, the instruction causes a general-protection exception. SDM 13.5.6". if enable afterkvm_put_guest_fpu()I think it too late.)
PATCH v2 8/8 KVM: x86: Disable intercept for Intel processor trace MSRs
In reply to Luwei KangFrom: Chao Peng <[email protected]>
Pass through Intel processor trace to guest directly.
--- arch/x86/kvm/vmx.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
arch/x86/kvm/vmx.c +13 -0
@@ -7029,6 +7029,19 @@ static __init int hardware_setup(void) !cpu_has_vmx_intel_pt() || !cpu_has_vmx_pt_use_gpa()) pt_mode =PT_MODE_SYSTEM ; +if (pt_mode ==PT_MODE_HOST_GUEST ) { +u32 i, eax, ebx, ecx, edx; + +cpuid_count (0x14 ,1 , &eax, &ebx, &ecx, &edx); +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_STATUS , false); +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_OUTPUT_BASE , false); +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_OUTPUT_MASK , false); +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_CR3_MATCH , false); +for (i =0 ; i < (eax &0x7 ); i++) +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_ADDR0_A + i, + false); + } +return alloc_kvm_area (); out:
Signature · 1 line
1.8.3.1
Paolo Bonzini
maintainerRe: [patch v2 8/8] KVM: x86: Disable intercept for Intel processor trace MSRs
Quoted 21 lines · Luwei Kang
From: Chao Peng <[email protected]>
Pass through Intel processor trace to guest directly.
Signed-off-by Chao Peng <[email protected]>Signed-off-by Luwei Kang <[email protected]>--- arch/x86/kvm/vmx.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)arch/x86/kvm/vmx.c +3
-0@@ -7029,6 +7029,19 @@ static __init int hardware_setup(void) !cpu_has_vmx_intel_pt() || !cpu_has_vmx_pt_use_gpa()) pt_mode =PT_MODE_SYSTEM ; +if (pt_mode ==PT_MODE_HOST_GUEST ) { +u32 i, eax, ebx, ecx, edx; +
arch/x86/kvm/vmx.c static __init int hardware_setup(void) · line 7044
Quoted 10 lines
arch/x86/kvm/vmx.c +10
-0+cpuid_count (0x14 ,1 , &eax, &ebx, &ecx, &edx); +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_STATUS , false); +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_OUTPUT_BASE , false); +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_OUTPUT_MASK , false); +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_CR3_MATCH , false); +for (i =0 ; i < (eax &0x7 ); i++) +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_ADDR0_A + i, + false); + } +
As I mentioned earlier, this probably makes vmentry/vmexit too expensive when guests are not using processor tracing. I would do it only if guest TRACEEN=1 (since anyway the values have to be correct if guest TRACEEN=1, and a change in TRACEEN always causes a vmexit).
Paolo
Quoted 4 lines
arch/x86/kvm/vmx.c +0
-0return alloc_kvm_area (); out:
RE: [patch v2 8/8] KVM: x86: Disable intercept for Intel processor trace MSRs
Quoted 7 lines
Quoted 7 lines
arch/x86/kvm/vmx.c +7
-0+if (pt_mode ==PT_MODE_HOST_GUEST ) { +u32 i, eax, ebx, ecx, edx; + +cpuid_count (0x14 ,1 , &eax, &ebx, &ecx, &edx); +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_STATUS , false); +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_OUTPUT_BASE , false); +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_OUTPUT_MASK , false);
Quoted 9 lines
arch/x86/kvm/vmx.c static __init int hardware_setup(void) · line 7044
Quoted 6 lines
arch/x86/kvm/vmx.c +6
-0+vmx_disable_intercept_for_msr (MSR_IA32_RTIT_CR3_MATCH , false); +for (i =0 ; i < (eax &0x7 ); i++) +vmx_disable_intercept_for_msr (MSR_IA32_RTIT_ADDR0_A + i, + false); + } +As I mentioned earlier, this probably makes vmentry/vmexit too expensive when guests are not using processor tracing. I would do it only if guest TRACEEN=1 (since anyway the values have to be correct if guest TRACEEN=1, and a change in TRACEEN always causes a vmexit).
Will change in next version.
Thanks,
Luwei Kang
Quoted 5 lines
Quoted 5 lines
return
alloc_kvm_area();out:
Paolo Bonzini
maintainerQuoted 4 lines · Luwei Kang
From v1:
- remove guest-only mode because guest-only mode can be covered by host-guest mode;
- always set "use GPA for processor tracing" in secondary execution control if it can be;
- trapRTIT_CTLread/write. Forbid write this msr when VMXON in L1 hypervisor.
Sounds fine. However, it's likely that this will only be in Linux 4.16.
Paolo
Quoted 7 lines
Quoted 4 lines
From v1:
- remove guest-only mode because guest-only mode can be covered by host-guest mode;
- always set "use GPA for processor tracing" in secondary execution control if it can be;
- trapRTIT_CTLread/write. Forbid write this msr when VMXON in L1 hypervisor.Sounds fine. However, it's likely that this will only be in Linux 4.16.
OK. Hope can get more comments to these patches. Thanks.
Luwei Kang