x86/split_lock: Fix and virtualization of split lock detection
v6 cover x86/split_lock: Fix and virtualization of split lock detection
So sorry for the noise that I forgot to CC the maillist.
This series aims to add the virtualization of split lock detection for guest, while containing some fixes of native kernel split lock handling.
Note, this series is based on the kernel patch[1]. Patch 1-3 are x86 kernel patches that based on the linux/master branch. Patch 4-8 are kvm patches that based on the kvm/queue branch.
Patch 1 is the fix and enhancement for kernel split lock detction. It ensures X86_FEATURE_SPLIT_LOCK_DETECT flag is set only when feature does exist and not disabled on kernel params. And it explicitly turn off split lock when sld_off instead of assuming BIOS/firmware leaves it cleared.
Patch 2 optimizes the runtime MSR accessing.
Patch 3 are the preparation for enabling split lock detection virtualization in KVM.
Patch 4 fixes the issue that malicious guest may exploit kvm emulator to attcact host kernel.
Patch 5 handles guest's split lock when host turns split lock detect on.
Patch 6-8 implement the virtualization of split lock detection in kvm.
[1]: https://lore.kernel.org/lkml/158031147976.396.8941798847364718785.tip-bot2@tip-bot2/
Changes in v6:
- Drop the sld_not_exist flag and use X86_FEATURE_SPLIT_LOCK_DETECT to check whether need to init split lock detection. [tglx]
- Use tglx's method to verify the existence of split lock detectoin.
- small optimization of sld_update_msr() that the default value of msr_test_ctrl_cache has split_lock_detect bit cleared.
- Drop the patch3 in v5 that introducing kvm_only option. [tglx]
- Rebase patch4-8 to kvm/queue.
- use the new kvm-cpu-cap to expose X86_FEATURE_CORE_CAPABILITIES in Patch 6.
Changes in v5:
- Use X86_FEATURE_SPLIT_LOCK_DETECT flag in kvm to ensure split lock detection is really supported.
- Add and export sld related helper functions in their related usecase kvm patches.
Changes in v4:
- Add patch 1 to rework the initialization flow of split lock detection.
- Drop percpu MSR_TEST_CTRL cache, just use a static variable to cache the reserved/unused bit of MSR_TEST_CTRL. [Sean]
- Add new option for split_lock_detect kernel param.
- Changlog refinement. [Sean]
- Add a new patch to enable MSR_TEST_CTRL for intel guest. [Sean]
Xiaoyao Li (8):
x86/split_lock: Rework the initialization flow of split lock detection x86/split_lock: Avoid runtime reads of the TEST_CTRL MSR x86/split_lock: Export handle_user_split_lock()
kvm: x86: Emulate split-lock access as a write in emulator
kvm: vmx: Extend VMX's #AC interceptor to handle split lock #AC
happens in guest
kvm: x86: Emulate MSR IA32_CORE_CAPABILITIES
kvm: vmx: Enable MSR_TEST_CTRL for intel guest
kvm: vmx: virtualize split lock detection
arch/x86/include/asm/cpu.h | 21 +++++- arch/x86/include/asm/kvm_host.h | 1 + arch/x86/kernel/cpu/intel.c | 114 +++++++++++++++++++------------- arch/x86/kernel/traps.c | 2 +- arch/x86/kvm/cpuid.c | 1 + arch/x86/kvm/vmx/vmx.c | 75 ++++++++++++++++++++- arch/x86/kvm/vmx/vmx.h | 1 + arch/x86/kvm/x86.c | 42 +++++++++++- 8 files changed, 203 insertions(+), 54 deletions(-)
Signature · 1 line
2.20.1
PATCH v6 1/8 x86/split_lock: Rework the initialization flow of split lock detection
Current initialization flow of split lock detection has following issues:
1. It assumes the initial value of MSR_TEST_CTRL.SPLIT_LOCK_DETECT to be zero. However, it's possible that BIOS/firmware has set it.
2. X86_FEATURE_SPLIT_LOCK_DETECT flag is unconditionally set even if there is a virtualization flaw that FMS indicates the existence while it's actually not supported.
3. Because of #2, for nest virt, L1 KVM cannot rely on flag X86_FEATURE_SPLIT_LOCK_DETECT to check the existence of feature.
Rework the initialization flow to solve above issues. In detail, explicitly set and clear split_lock_detect bit to verify MSR_TEST_CTRL can be accessed, and rdmsr after wrmsr to ensure bit is set successfully.
X86_FEATURE_SPLIT_LOCK_DETECT flag is set only when the feature does exist and the feature is not disabled with kernel param "split_lock_detect=off"
--- arch/x86/kernel/cpu/intel.c | 79 +++++++++++++++++++++---------------- 1 file changed, 46 insertions(+), 33 deletions(-)
arch/x86/kernel/cpu/intel.c +46 -33
@@ -44,7 +44,7 @@ enum split_lock_detect_state {* split_lock_setup() will switch this to sld_warn on systems that support * split lock detect, unless there is a command line override. */ -static enum split_lock_detect_state sld_state = sld_off; +static enum split_lock_detect_state sld_state__ro_after_init = sld_off;/* * Processors which have self-snooping capability can handle conflicting
@@ -984,78 +984,91 @@ static inline bool match_option(const char *arg, int arglen, const char *opt)return len == arglen && !strncmp (arg, opt, len); } +static bool __init split_lock_verify_msr (bool on) +{ +u64 ctrl, tmp; + +if (rdmsrl_safe (MSR_TEST_CTRL , &ctrl)) +return false; + +if (on) + ctrl |=MSR_TEST_CTRL_SPLIT_LOCK_DETECT ; +else + ctrl &= ~MSR_TEST_CTRL_SPLIT_LOCK_DETECT ; + +if (wrmsrl_safe (MSR_TEST_CTRL , ctrl)) +return false; + +rdmsrl (MSR_TEST_CTRL , tmp); +return ctrl == tmp; +} +static void __init split_lock_setup (void ) { +enum split_lock_detect_state state = sld_warn;char arg[20 ];int i, ret; -setup_force_cpu_cap (X86_FEATURE_SPLIT_LOCK_DETECT ); - sld_state = sld_warn; +if (!split_lock_verify_msr (false)) { +pr_info ("MSR access failed: Disabled\n" ); +return ; + } ret =cmdline_find_option (boot_command_line,"split_lock_detect" , arg,sizeof (arg));if (ret >=0 ) {for (i =0 ; i <ARRAY_SIZE (sld_options); i++) {if (match_option (arg, ret, sld_options[i].option)) { - sld_state = sld_options[i].state; + state = sld_options[i].state;break ; } } } -switch (sld_state) { +switch (state) {case sld_off:pr_info ("disabled\n" ); -break ; - +return ;case sld_warn:pr_info ("warning about user-space split_locks\n" );break ; -case sld_fatal:pr_info ("sending SIGBUS on user-space split_locks\n" );break ; } + +if (!split_lock_verify_msr (true)) { +pr_info ("MSR access failed: Disabled\n" ); +return ; + } + + sld_state = state; +setup_force_cpu_cap (X86_FEATURE_SPLIT_LOCK_DETECT ); }/* -* Locking is not required at the moment because only bit 29 of this -* MSR is implemented and locking would not prevent that the operation -* of one thread is immediately undone by the sibling thread. -* Use the "safe" versions of rdmsr/wrmsr here because although code -* checks CPUID and MSR bits to make sure the TEST_CTRL MSR should -* exist, there may be glitches in virtualization that leave a guest -* with an incorrect view of real h/w capabilities. +* MSR_TEST_CTRL is per core, but we treat it like a per CPU MSR. Locking +* is not implemented as one thread could undo the setting of the other +* thread immediately after dropping the lock anyway. */ -static bool __sld_msr_set (bool on) +static void sld_update_msr (bool on) {u64 test_ctrl_val; -if (rdmsrl_safe (MSR_TEST_CTRL , &test_ctrl_val)) -return false; +rdmsrl (MSR_TEST_CTRL , test_ctrl_val);if (on) test_ctrl_val |=MSR_TEST_CTRL_SPLIT_LOCK_DETECT ;else test_ctrl_val &= ~MSR_TEST_CTRL_SPLIT_LOCK_DETECT ; -return !wrmsrl_safe (MSR_TEST_CTRL , test_ctrl_val); +wrmsrl (MSR_TEST_CTRL , test_ctrl_val); }static void split_lock_init (void ) { -if (sld_state == sld_off) -return ; - -if (__sld_msr_set (true)) -return ; - -/* -* If this is anything other than the boot-cpu, you've done -* funny things and you get to keep whatever pieces. -*/ -pr_warn ("MSR fail -- disabled\n" ); - sld_state = sld_off; +if (boot_cpu_has (X86_FEATURE_SPLIT_LOCK_DETECT )) +sld_update_msr (sld_state != sld_off); }bool handle_user_split_lock (struct pt_regs *regs,long error_code)
@@ -1071,7 +1084,7 @@ bool handle_user_split_lock(struct pt_regs *regs, long error_code)* progress and set TIF_SLD so the detection is re-enabled via * switch_to_sld() when the task is scheduled out. */ -__sld_msr_set (false); +sld_update_msr (false);set_tsk_thread_flag (current,TIF_SLD );return true; }
@@ -1085,7 +1098,7 @@ bool handle_user_split_lock(struct pt_regs *regs, long error_code)*/ void switch_to_sld (unsigned long tifn) { -__sld_msr_set (!(tifn & _TIF_SLD)); +sld_update_msr (!(tifn & _TIF_SLD)); }#define SPLIT_LOCK_CPU (model) {X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY}
Signature · 1 line
2.20.1
PATCH v6 2/8 x86/split_lock: Avoid runtime reads of the TEST_CTRL MSR
In reply to Xiaoyao LiIn a context switch from a task that is detecting split locks to one that is not (or vice versa) we need to update the TEST_CTRL MSR. Currently this is done with the common sequence:
read the MSR
flip the bit
write the MSR
in order to avoid changing the value of any reserved bits in the MSR.
Cache unused and reserved bits of TEST_CTRL MSR with SPLIT_LOCK_DETECT bit cleared during initialization, so we can avoid an expensive RDMSR instruction during context switch.
--- arch/x86/kernel/cpu/intel.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-)
arch/x86/kernel/cpu/intel.c +4 -5
@@ -45,6 +45,7 @@ enum split_lock_detect_state {* split lock detect, unless there is a command line override. */ static enum split_lock_detect_state sld_state__ro_after_init = sld_off; +static u64 msr_test_ctrl_cache__ro_after_init ;/* * Processors which have self-snooping capability can handle conflicting
@@ -1037,6 +1038,8 @@ static void __init split_lock_setup(void)break ; } +rdmsrl (MSR_TEST_CTRL , msr_test_ctrl_cache); +if (!split_lock_verify_msr (true)) {pr_info ("MSR access failed: Disabled\n" );return ;
@@ -1053,14 +1056,10 @@ static void __init split_lock_setup(void)*/ static void sld_update_msr (bool on) { -u64 test_ctrl_val; - -rdmsrl (MSR_TEST_CTRL , test_ctrl_val); +u64 test_ctrl_val = msr_test_ctrl_cache;if (on) test_ctrl_val |=MSR_TEST_CTRL_SPLIT_LOCK_DETECT ; -else - test_ctrl_val &= ~MSR_TEST_CTRL_SPLIT_LOCK_DETECT ;wrmsrl (MSR_TEST_CTRL , test_ctrl_val); }
Signature · 1 line
2.20.1
PATCH v6 3/8 x86/split_lock: Export handle_user_split_lock()
In reply to Xiaoyao LiIn the future, KVM will use handle_user_split_lock() to handle #AC caused by split lock in guest. Due to the fact that KVM doesn't have a @regs context and will pre-check EFLASG.AC, move the EFLAGS.AC check to do_alignment_check().
--- arch/x86/include/asm/cpu.h | 4 ++-- arch/x86/kernel/cpu/intel.c | 7 ++++--- arch/x86/kernel/traps.c | 2 +- 3 files changed, 7 insertions(+), 6 deletions(-)
arch/x86/include/asm/cpu.h +2 -2
@@ -43,11 +43,11 @@ unsigned int x86_stepping(unsigned int sig);#ifdef CONFIG_CPU_SUP_INTEL extern void __init cpu_set_core_cap_bits (struct cpuinfo_x86 *c);extern void switch_to_sld (unsigned long tifn); -extern bool handle_user_split_lock (struct pt_regs *regs,long error_code); +extern bool handle_user_split_lock (unsigned long ip);#else static inline void __init cpu_set_core_cap_bits (struct cpuinfo_x86 *c) {}static inline void switch_to_sld (unsigned long tifn) {} -static inline bool handle_user_split_lock (struct pt_regs *regs,long error_code) +static inline bool handle_user_split_lock (unsigned long ip) {return false; }
arch/x86/kernel/cpu/intel.c +4 -3
@@ -1070,13 +1070,13 @@ static void split_lock_init(void)sld_update_msr (sld_state != sld_off); } -bool handle_user_split_lock (struct pt_regs *regs,long error_code) +bool handle_user_split_lock (unsigned long ip) { -if ((regs->flags &X86_EFLAGS_AC ) || sld_state == sld_fatal) +if (sld_state == sld_fatal)return false;pr_warn_ratelimited ("#AC: %s/%d took a split_lock trap at address: 0x%lx\n" , - current->comm, current->pid, regs->ip); + current->comm, current->pid, ip);/* * Disable the split lock detection for this task so it can make
@@ -1087,6 +1087,7 @@ bool handle_user_split_lock(struct pt_regs *regs, long error_code)set_tsk_thread_flag (current,TIF_SLD );return true; } +EXPORT_SYMBOL_GPL (handle_user_split_lock);/* * This function is called only when switching between tasks with
arch/x86/kernel/traps.c +1 -1
@@ -304,7 +304,7 @@ dotraplinkage void do_alignment_check(struct pt_regs *regs, long error_code)local_irq_enable (); -if (handle_user_split_lock (regs, error_code)) +if (!(regs->flags &X86_EFLAGS_AC ) &&handle_user_split_lock (regs->ip))return ; do_trap(X86_TRAP_AC, SIGBUS, "alignment check", regs,
Signature · 1 line
2.20.1
PATCH v6 4/8 kvm: x86: Emulate split-lock access as a write in emulator
In reply to Xiaoyao LiIf split lock detect is on (warn/fatal), #AC handler calls die() when split lock happens in kernel.
Malicous guest can exploit the KVM emulator to trigger split lock #AC in kernel[1]. So just emulating the access as a write if it's a split-lock access (the same as access spans page) to avoid malicious guest attacking kernel.
More discussion can be found [2][3].
[1] https://lore.kernel.org/lkml/[email protected]/
[2] https://lkml.kernel.org/r/[email protected]
[3] https://lkml.kernel.org/r/[email protected]
--- arch/x86/include/asm/cpu.h | 2 ++ arch/x86/kernel/cpu/intel.c | 6 ++++++ arch/x86/kvm/x86.c | 7 ++++++- 3 files changed, 14 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpu.h +2 -0
@@ -44,6 +44,7 @@ unsigned int x86_stepping(unsigned int sig);extern void __init cpu_set_core_cap_bits (struct cpuinfo_x86 *c);extern void switch_to_sld (unsigned long tifn);extern bool handle_user_split_lock (unsigned long ip); +extern bool split_lock_detect_on (void );#else static inline void __init cpu_set_core_cap_bits (struct cpuinfo_x86 *c) {}static inline void switch_to_sld (unsigned long tifn) {}
@@ -51,5 +52,6 @@ static inline bool handle_user_split_lock(unsigned long ip) {return false; } +static inline bool split_lock_detect_on (void ) {return false; }#endif #endif /* _ASM_X86_CPU_H */
arch/x86/kernel/cpu/intel.c +6 -0
@@ -1070,6 +1070,12 @@ static void split_lock_init(void)sld_update_msr (sld_state != sld_off); } +bool split_lock_detect_on (void ) +{ +return sld_state != sld_off; +} +EXPORT_SYMBOL_GPL (split_lock_detect_on); +bool handle_user_split_lock (unsigned long ip) {if (sld_state == sld_fatal)
arch/x86/kvm/x86.c +6 -1
@@ -5831,6 +5831,7 @@ static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt, {struct kvm_host_map map;struct kvm_vcpu *vcpu =emul_to_vcpu (ctxt); +u64 page_line_mask =PAGE_MASK ;gpa_t gpa;char *kaddr;bool exchanged;
@@ -5845,7 +5846,11 @@ static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt, (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) goto emul_write; -if (((gpa + bytes -1 ) &PAGE_MASK ) != (gpa &PAGE_MASK )) +if (split_lock_detect_on ()) + page_line_mask = ~(cache_line_size () -1 ); + +/* when write spans page or spans cache when SLD enabled */ +if (((gpa + bytes -1 ) & page_line_mask) != (gpa & page_line_mask)) goto emul_write;if (kvm_vcpu_map (vcpu,gpa_to_gfn (gpa), &map))
Signature · 1 line
2.20.1
Re: [PATCH v6 4/8] kvm: x86: Emulate split-lock access as a write in emulator
arch/x86/kernel/cpu/intel.c static void split_lock_init(void) · line 1077
Quoted 6 lines · Xiaoyao Li
arch/x86/kernel/cpu/intel.c +5
-0+bool split_lock_detect_on (void ) +{ +return sld_state != sld_off; +} +EXPORT_SYMBOL_GPL (split_lock_detect_on);
1) You export this function here
2) You change that in one of the next patches to something else
3) According to patch 1/8 X86_FEATURE_SPLIT_LOCK_DETECT is not set when sld_state == sld_off. FYI, I did that on purpose.
AFAICT #1 and #2 are just historical leftovers of your previous patch series and the extra step was just adding more changed lines per patch for no value.
#3 changed the detection mechanism and at the same time the semantics of the feature flag.
So what's the point of this exercise?
Thanks,
tglx
Re: [PATCH v6 4/8] kvm: x86: Emulate split-lock access as a write in emulator
Quoted 15 lines · Thomas Gleixner
arch/x86/kernel/cpu/intel.c static void split_lock_init(void) · line 1077
Quoted 6 lines · Xiaoyao Li
arch/x86/kernel/cpu/intel.c +5
-0+bool split_lock_detect_on (void ) +{ +return sld_state != sld_off; +} +EXPORT_SYMBOL_GPL (split_lock_detect_on);1) You export this function here
2) You change that in one of the next patches to something else
3) According to patch 1/8
X86_FEATURE_SPLIT_LOCK_DETECTis not set whensld_state == sld_off. FYI, I did that on purpose.
Quoted 8 lines
AFAICT #1 and #2 are just historical leftovers of your previous patch series and the extra step was just adding more changed lines per patch for no value.
#3 changed the detection mechanism and at the same time the semantics of the feature flag.
So what's the point of this exercise?
Right. In this series, setting X86_FEATURE_SPLIT_LOCK_DETECT flag means SLD is turned on. Need to remove split_lock_detect_on(). Thanks for pointing out this.
Quoted 4 lines
Thanks,
tglx
PATCH v6 5/8 kvm: vmx: Extend VMX's #AC interceptor to handle split lock #AC happens in guest
In reply to Xiaoyao LiThere are two types of #AC can be generated in Intel CPUs:
1. legacy alignment check #AC;
2. split lock #AC;
Legacy alignment check #AC can be injected to guest if guest has enabled alignemnet check.
when host enables split lock detectin, i.e., sld_warn or sld_fatal, there will be an unexpected #AC in guest and intercepted by KVM because KVM doesn't virtualize this feature to guest and hardware value of MSR_TEST_CTRL.SLD bit stays unchanged when vcpu is running.
To handle this unexpected #AC, treat guest just like host usermode that calling handle_user_split_lock():
- If host is sld_warn, it warns and set TIF_SLD so that __switch_to_xtra() does the MSR_TEST_CTRL.SLD bit switching when control transfer to/from this vcpu.
- If host is sld_fatal, forward #AC to userspace, the similar as sending SIGBUS.
--- arch/x86/kvm/vmx/vmx.c | 30 +++++++++++++++++++++++++++--- 1 file changed, 27 insertions(+), 3 deletions(-)
arch/x86/kvm/vmx/vmx.c +27 -3
@@ -4613,6 +4613,12 @@ static int handle_machine_check(struct kvm_vcpu *vcpu)return 1 ; } +static inline bool guest_cpu_alignment_check_enabled (struct kvm_vcpu *vcpu) +{ +return vmx_get_cpl (vcpu) ==3 &&kvm_read_cr0_bits (vcpu,X86_CR0_AM ) && + (kvm_get_rflags (vcpu) &X86_EFLAGS_AC ); +} +static int handle_exception_nmi (struct kvm_vcpu *vcpu) {struct vcpu_vmx *vmx =to_vmx (vcpu);
@@ -4678,9 +4684,6 @@ static int handle_exception_nmi(struct kvm_vcpu *vcpu)return handle_rmode_exception (vcpu, ex_no, error_code);switch (ex_no) { -case AC_VECTOR : -kvm_queue_exception_e (vcpu,AC_VECTOR , error_code); -return 1 ;case DB_VECTOR : dr6 =vmcs_readl (EXIT_QUALIFICATION ); if (!(vcpu->guest_debug &
@@ -4709,6 +4712,27 @@ static int handle_exception_nmi(struct kvm_vcpu *vcpu) kvm_run->debug.arch.pc =vmcs_readl (GUEST_CS_BASE ) + rip; kvm_run->debug.arch.exception = ex_no;break ; +case AC_VECTOR : +/* +* Reflect #AC to the guest if it's expecting the #AC, i.e. has +* legacy alignment check enabled. Pre-check host split lock +* support to avoid the VMREADs needed to check legacy #AC, +* i.e. reflect the #AC if the only possible source is legacy +* alignment checks. +*/ +if (!split_lock_detect_on () || +guest_cpu_alignment_check_enabled (vcpu)) { +kvm_queue_exception_e (vcpu,AC_VECTOR , error_code); +return 1 ; + } + +/* +* Forward the #AC to userspace if kernel policy does not allow +* temporarily disabling split lock detection. +*/ +if (handle_user_split_lock (kvm_rip_read (vcpu))) +return 1 ; + fallthrough;default : kvm_run->exit_reason =KVM_EXIT_EXCEPTION ; kvm_run->ex.exception = ex_no;
Signature · 1 line
2.20.1
PATCH v6 6/8 kvm: x86: Emulate MSR IA32_CORE_CAPABILITIES
In reply to Xiaoyao LiEmulate MSR_IA32_CORE_CAPABILITIES in software and unconditionally advertise its support to userspace. Like MSR_IA32_ARCH_CAPABILITIES, it is a feature-enumerating MSR and can be fully emulated regardless of hardware support. Existence of CORE_CAPABILITIES is enumerated via CPUID.(EAX=7H,ECX=0):EDX[30].
Note, support for individual features enumerated via CORE_CAPABILITIES, e.g., split lock detection, will be added in future patches.
--- arch/x86/include/asm/kvm_host.h | 1 + arch/x86/kvm/cpuid.c | 1 + arch/x86/kvm/x86.c | 22 ++++++++++++++++++++++ 3 files changed, 24 insertions(+)
arch/x86/include/asm/kvm_host.h +1 -0
@@ -597,6 +597,7 @@ struct kvm_vcpu_arch {u64 ia32_xss;u64 microcode_version;u64 arch_capabilities; +u64 core_capabilities;/* * Paging state of the vcpu
arch/x86/kvm/cpuid.c +1 -0
@@ -344,6 +344,7 @@ void kvm_set_cpu_caps(void)/* TSC_ADJUST and ARCH_CAPABILITIES are emulated in software. */ kvm_cpu_cap_set (X86_FEATURE_TSC_ADJUST );kvm_cpu_cap_set (X86_FEATURE_ARCH_CAPABILITIES ); +kvm_cpu_cap_set (X86_FEATURE_CORE_CAPABILITIES );if (boot_cpu_has (X86_FEATURE_IBPB ) &&boot_cpu_has (X86_FEATURE_IBRS ))kvm_cpu_cap_set (X86_FEATURE_SPEC_CTRL );
arch/x86/kvm/x86.c +22 -0
@@ -1248,6 +1248,7 @@ static const u32 emulated_msrs_all[] = {MSR_IA32_TSC_ADJUST ,MSR_IA32_TSCDEADLINE ,MSR_IA32_ARCH_CAPABILITIES , +MSR_IA32_CORE_CAPS ,MSR_IA32_MISC_ENABLE ,MSR_IA32_MCG_STATUS , MSR_IA32_MCG_CTL,
@@ -1314,6 +1315,7 @@ static const u32 msr_based_features_all[] = {MSR_F10H_DECFG ,MSR_IA32_UCODE_REV ,MSR_IA32_ARCH_CAPABILITIES , + MSR_IA32_CORE_CAPS, };static u32 msr_based_features[ARRAY_SIZE (msr_based_features_all)];
@@ -1367,12 +1369,20 @@ static u64 kvm_get_arch_capabilities(void)return data; } +static u64 kvm_get_core_capabilities (void ) +{ +return 0 ; +} +static int kvm_get_msr_feature (struct kvm_msr_entry *msr) {switch (msr->index) {case MSR_IA32_ARCH_CAPABILITIES : msr->data =kvm_get_arch_capabilities ();break ; +case MSR_IA32_CORE_CAPS : + msr->data =kvm_get_core_capabilities (); +break ;case MSR_IA32_UCODE_REV :rdmsrl_safe (msr->index, &msr->data);break ;
@@ -2745,6 +2755,11 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)return 1 ; vcpu->arch.arch_capabilities = data;break ; +case MSR_IA32_CORE_CAPS : +if (!msr_info->host_initiated) +return 1 ; + vcpu->arch.core_capabilities = data; +break ;case MSR_EFER :return set_efer (vcpu, msr_info);case MSR_K7_HWCR :
@@ -3072,6 +3087,12 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)return 1 ; msr_info->data = vcpu->arch.arch_capabilities;break ; +case MSR_IA32_CORE_CAPS : +if (!msr_info->host_initiated && + !guest_cpuid_has (vcpu,X86_FEATURE_CORE_CAPABILITIES )) +return 1 ; + msr_info->data = vcpu->arch.core_capabilities; +break ;case MSR_IA32_POWER_CTL : msr_info->data = vcpu->arch.msr_ia32_power_ctl;break ;
@@ -9367,6 +9388,7 @@ int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) goto free_guest_fpu; vcpu->arch.arch_capabilities =kvm_get_arch_capabilities (); + vcpu->arch.core_capabilities =kvm_get_core_capabilities (); vcpu->arch.msr_platform_info =MSR_PLATFORM_INFO_CPUID_FAULT ;kvm_vcpu_mtrr_init (vcpu);vcpu_load (vcpu);
Signature · 1 line
2.20.1
PATCH v6 7/8 kvm: vmx: Enable MSR_TEST_CTRL for intel guest
In reply to Xiaoyao LiOnly enabling the read and write zero of MSR_TEST_CTRL. This makes MSR_TEST_CTRL always available for intel guest, but guset cannot write any value to it except zero.
This matches the truth that most Intel CPUs support MSR_TEST_CTRL, and it also alleviates the effort to handle wrmsr/rdmsr when exposing split lock detect to guest in the following patch.
--- arch/x86/kvm/vmx/vmx.c | 10 ++++++++++ arch/x86/kvm/vmx/vmx.h | 1 + 2 files changed, 11 insertions(+)
arch/x86/kvm/vmx/vmx.c +10 -0
@@ -1820,6 +1820,9 @@ static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)u32 index;switch (msr_info->index) { +case MSR_TEST_CTRL : + msr_info->data = vmx->msr_test_ctrl; +break ;#ifdef CONFIG_X86_64 case MSR_FS_BASE : msr_info->data =vmcs_readl (GUEST_FS_BASE );
@@ -1973,6 +1976,12 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)u32 index;switch (msr_index) { +case MSR_TEST_CTRL : +if (data) +return 1 ; + + vmx->msr_test_ctrl = data; +break ;case MSR_EFER : ret =kvm_set_msr_common (vcpu, msr_info);break ;
@@ -4283,6 +4292,7 @@ static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) vmx->rmode.vm86_active =0 ; vmx->spec_ctrl =0 ; + vmx->msr_test_ctrl =0 ; vmx->msr_ia32_umwait_control =0 ;
arch/x86/kvm/vmx/vmx.h +1 -0
@@ -224,6 +224,7 @@ struct vcpu_vmx {#endif u64 spec_ctrl; +u64 msr_test_ctrl;u32 msr_ia32_umwait_control;u32 secondary_exec_control;
Signature · 1 line
2.20.1
Re: [PATCH v6 7/8] kvm: vmx: Enable MSR_TEST_CTRL for intel guest
Quoted 1 line · Xiaoyao Li
Subject: Re: [PATCH v6 7/8] kvm: vmx: Enable
MSR_TEST_CTRLfor intel guest
What the heck is a "intel guest"?
Can you Intel folks please stop to slap Intel (and in your subject line it's even spelled wrong) at everything whether it makes sense or not?
Thanks,
tglx
PATCH v6 8/8 kvm: vmx: virtualize split lock detection
In reply to Xiaoyao LiDue to the fact that MSR_TEST_CTRL is per-core scope, i.e., the sibling threads in the same physical CPU core share the same MSR, only advertising feature split lock detection to guest when SMT is disabled or unsupported, for simplicitly.
Below summarizing how guest behaves of different host configuration:
sld_fatal - hardware MSR_TEST_CTRL.SLD is always on when vcpu is running,
even though guest thinks it sets/clears MSR_TEST_CTRL.SLD bit successfully. i.e., SLD is forced on for guest.
sld_warn - hardware MSR_TEST_CTRL.SLD is left on until an #AC is
intercepted with MSR_TEST_CTRL.SLD=0 in the guest, at which point normal sld_warn rules apply, i.e., clear MSR_TEST_CTRL.SLD bit and set TIF_SLD.
If a vCPU associated with the task does VM-Enter with virtual MSR_TEST_CTRL.SLD=1, TIF_SLD is reset, hardware MSR_TEST_CTRL.SLD is re-set, and cycle begins anew.
sld_disable - guest cannot see feature split lock detection.
--- arch/x86/include/asm/cpu.h | 17 +++++++++++++- arch/x86/kernel/cpu/intel.c | 29 +++++++++++++----------- arch/x86/kvm/vmx/vmx.c | 45 ++++++++++++++++++++++++++++++++----- arch/x86/kvm/x86.c | 17 +++++++++++--- 4 files changed, 86 insertions(+), 22 deletions(-)
arch/x86/include/asm/cpu.h +16 -1
@@ -41,10 +41,23 @@ unsigned int x86_family(unsigned int sig);unsigned int x86_model (unsigned int sig);unsigned int x86_stepping (unsigned int sig);#ifdef CONFIG_CPU_SUP_INTEL +enum split_lock_detect_state { + sld_off =0 , + sld_warn, + sld_fatal, +}; +extern enum split_lock_detect_state sld_state__ro_after_init ; + +static inline bool split_lock_detect_on (void ) +{ +return sld_state != sld_off; +} +extern void __init cpu_set_core_cap_bits (struct cpuinfo_x86 *c);extern void switch_to_sld (unsigned long tifn);extern bool handle_user_split_lock (unsigned long ip); -extern bool split_lock_detect_on (void ); +extern void sld_msr_set (bool on); +extern void sld_turn_back_on (void );#else static inline void __init cpu_set_core_cap_bits (struct cpuinfo_x86 *c) {}static inline void switch_to_sld (unsigned long tifn) {}
@@ -53,5 +66,7 @@ static inline bool handle_user_split_lock(unsigned long ip)return false; }static inline bool split_lock_detect_on (void ) {return false; } +static inline void sld_msr_set (bool on) {} +static inline void sld_turn_back_on (void ) {}#endif #endif /* _ASM_X86_CPU_H */
arch/x86/kernel/cpu/intel.c +16 -13
@@ -33,18 +33,14 @@#include <asm/apic.h> #endif -enum split_lock_detect_state { - sld_off =0 , - sld_warn, - sld_fatal, -}; -/* * Default to sld_off because most systems do not support split lock detection * split_lock_setup() will switch this to sld_warn on systems that support * split lock detect, unless there is a command line override. */ -static enum split_lock_detect_state sld_state__ro_after_init = sld_off; +enum split_lock_detect_state sld_state__ro_after_init = sld_off; +EXPORT_SYMBOL_GPL (sld_state); +static u64 msr_test_ctrl_cache__ro_after_init ;/*
@@ -1070,12 +1066,6 @@ static void split_lock_init(void)sld_update_msr (sld_state != sld_off); } -bool split_lock_detect_on (void ) -{ -return sld_state != sld_off; -} -EXPORT_SYMBOL_GPL (split_lock_detect_on); -bool handle_user_split_lock (unsigned long ip) {if (sld_state == sld_fatal)
@@ -1095,6 +1085,19 @@ bool handle_user_split_lock(unsigned long ip) }EXPORT_SYMBOL_GPL (handle_user_split_lock); +void sld_msr_set (bool on) +{ +sld_update_msr (on); +} +EXPORT_SYMBOL_GPL (sld_msr_set); + +void sld_turn_back_on (void ) +{ +sld_update_msr (true); +clear_tsk_thread_flag (current,TIF_SLD ); +} +EXPORT_SYMBOL_GPL (sld_turn_back_on); +/* * This function is called only when switching between tasks with * different split-lock detection modes. It sets the MSR for the
arch/x86/kvm/vmx/vmx.c +40 -5
83 lines not on this page · raw message
arch/x86/kvm/x86.c +14 -3
@@ -1189,7 +1189,7 @@ static const u32 msrs_to_save_all[] = {#endif MSR_IA32_TSC ,MSR_IA32_CR_PAT ,MSR_VM_HSAVE_PA ,MSR_IA32_FEAT_CTL ,MSR_IA32_BNDCFGS ,MSR_TSC_AUX , -MSR_IA32_SPEC_CTRL , +MSR_IA32_SPEC_CTRL ,MSR_TEST_CTRL ,MSR_IA32_RTIT_CTL ,MSR_IA32_RTIT_STATUS ,MSR_IA32_RTIT_CR3_MATCH ,MSR_IA32_RTIT_OUTPUT_BASE ,MSR_IA32_RTIT_OUTPUT_MASK , MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
@@ -1371,7 +1371,12 @@ static u64 kvm_get_arch_capabilities(void)static u64 kvm_get_core_capabilities (void ) { -return 0 ; +u64 data =0 ; + +if (boot_cpu_has (X86_FEATURE_SPLIT_LOCK_DETECT ) && !cpu_smt_possible ()) + data |=MSR_IA32_CORE_CAPS_SPLIT_LOCK_DETECT ; + +return data; }static int kvm_get_msr_feature (struct kvm_msr_entry *msr)
@@ -2756,7 +2761,8 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) vcpu->arch.arch_capabilities = data;break ;case MSR_IA32_CORE_CAPS : -if (!msr_info->host_initiated) +if (!msr_info->host_initiated || + data & ~kvm_get_core_capabilities ())return 1 ; vcpu->arch.core_capabilities = data;break ;
@@ -5235,6 +5241,11 @@ static void kvm_init_msr_list(void)* to the guests in some cases. */ switch (msrs_to_save_all[i]) { +case MSR_TEST_CTRL : +if (!(kvm_get_core_capabilities () & +MSR_IA32_CORE_CAPS_SPLIT_LOCK_DETECT )) +continue ; +break ;case MSR_IA32_BNDCFGS :if (!kvm_mpx_supported ())continue ;
Signature · 1 line
2.20.1
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
arch/x86/include/asm/cpu.h unsigned int x86_family(unsigned int sig); · line 54
Quoted 12 lines · Xiaoyao Li
#ifdef
CONFIG_CPU_SUP_INTELarch/x86/include/asm/cpu.h +11
-0+enum split_lock_detect_state { + sld_off =0 , + sld_warn, + sld_fatal, +}; +extern enum split_lock_detect_state sld_state__ro_after_init ; + +static inline bool split_lock_detect_on (void ) +{ +return sld_state != sld_off; +}
See previous reply.
arch/x86/kernel/cpu/intel.c bool handle_user_split_lock(unsigned long ip) · line 1099
Quoted 12 lines
arch/x86/kernel/cpu/intel.c +12
-0+void sld_msr_set (bool on) +{ +sld_update_msr (on); +} +EXPORT_SYMBOL_GPL (sld_msr_set); + +void sld_turn_back_on (void ) +{ +sld_update_msr (true); +clear_tsk_thread_flag (current,TIF_SLD ); +} +EXPORT_SYMBOL_GPL (sld_turn_back_on);
First of all these functions want to be in a separate patch, but aside of that they do not make any sense at all.
Quoted 14 lines
arch/x86/kvm/vmx/vmx.c +5
-2+static inline bool guest_cpu_split_lock_detect_on (struct vcpu_vmx *vmx) +{ +return vmx->msr_test_ctrl &MSR_TEST_CTRL_SPLIT_LOCK_DETECT ; +} +static int handle_exception_nmi (struct kvm_vcpu *vcpu) {struct vcpu_vmx *vmx =to_vmx (vcpu);@@ -4725,12 +4746,13 @@ static int handle_exception_nmi(struct kvm_vcpu *vcpu)case AC_VECTOR :/* * Reflect #AC to the guest if it's expecting the #AC, i.e. has -* legacy alignment check enabled. Pre-check host split lock -* support to avoid the VMREADs needed to check legacy #AC,
arch/x86/kvm/vmx/vmx.c static int handle_exception_nmi(struct kvm_vcpu *vcpu) · line 4756
Quoted 10 lines
arch/x86/kvm/vmx/vmx.c +5
-2-* i.e. reflect the #AC if the only possible source is legacy -* alignment checks. +* legacy alignment check enabled or split lock detect enabled. +* Pre-check host split lock support to avoid further check of +* guest, i.e. reflect the #AC if host doesn't enable split lock +* detection. */ if (!split_lock_detect_on () || +guest_cpu_split_lock_detect_on (vmx) ||guest_cpu_alignment_check_enabled (vcpu)) {
If the host has split lock detection disabled then how is the guest supposed to have it enabled in the first place?
arch/x86/kvm/vmx/vmx.c static void vmx_vcpu_run(struct kvm_vcpu *vcpu) · line 6659
Quoted 8 lines
arch/x86/kvm/vmx/vmx.c +4
-0@@ -6631,6 +6653,14 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu)*/ x86_spec_ctrl_set_guest (vmx->spec_ctrl,0 ); +if (static_cpu_has (X86_FEATURE_SPLIT_LOCK_DETECT ) && +guest_cpu_split_lock_detect_on (vmx)) { +if (test_thread_flag (TIF_SLD )) +sld_turn_back_on ();
This is completely inconsistent behaviour. The only way that TIF_SLD is set is when the host has sld_state == sld_warn and the guest triggered a split lock #AC.
'warn' means that the split lock event is registered and a printk emitted and after that the task runs with split lock detection disabled.
It does not matter at all if the task triggered the #AC while in guest or in host user space mode. Stop claiming that virt is special. The only special thing about virt is, that it is using a different mechanism to exit kernel mode. Aside of that from the kernel POV it is completely irrelevant whether the task triggered the split lock in host user space or in guest mode.
If the SLD mode is fatal, then the task is killed no matter what.
Please sit down and go through your patches and rethink every single line instead of sending out yet another half baken and hastily cobbled together pile.
To be clear, Patch 1 and 2 make sense on their own, so I'm tempted to pick them up right now, but the rest is going to be 5.8 material no matter what.
Thanks,
tglx
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
Quoted 51 lines · Thomas Gleixner
arch/x86/include/asm/cpu.h unsigned int x86_family(unsigned int sig); · line 54
Quoted 12 lines · Xiaoyao Li
#ifdef CONFIG_CPU_SUP_INTELarch/x86/include/asm/cpu.h +11
-0+enum split_lock_detect_state { + sld_off =0 , + sld_warn, + sld_fatal, +}; +extern enum split_lock_detect_state sld_state__ro_after_init ; + +static inline bool split_lock_detect_on (void ) +{ +return sld_state != sld_off; +}See previous reply.
arch/x86/kernel/cpu/intel.c bool handle_user_split_lock(unsigned long ip) · line 1099
Quoted 12 lines
arch/x86/kernel/cpu/intel.c +12
-0+void sld_msr_set (bool on) +{ +sld_update_msr (on); +} +EXPORT_SYMBOL_GPL (sld_msr_set); + +void sld_turn_back_on (void ) +{ +sld_update_msr (true); +clear_tsk_thread_flag (current,TIF_SLD ); +} +EXPORT_SYMBOL_GPL (sld_turn_back_on);First of all these functions want to be in a separate patch, but aside of that they do not make any sense at all.
Quoted 19 lines
arch/x86/kvm/vmx/vmx.c +8
-4+static inline bool guest_cpu_split_lock_detect_on (struct vcpu_vmx *vmx) +{ +return vmx->msr_test_ctrl &MSR_TEST_CTRL_SPLIT_LOCK_DETECT ; +} +static int handle_exception_nmi (struct kvm_vcpu *vcpu) {struct vcpu_vmx *vmx =to_vmx (vcpu);@@ -4725,12 +4746,13 @@ static int handle_exception_nmi(struct kvm_vcpu *vcpu)case AC_VECTOR :/* * Reflect #AC to the guest if it's expecting the #AC, i.e. has -* legacy alignment check enabled. Pre-check host split lock -* support to avoid the VMREADs needed to check legacy #AC, -* i.e. reflect the #AC if the only possible source is legacy -* alignment checks. +* legacy alignment check enabled or split lock detect enabled. +* Pre-check host split lock support to avoid further check of +* guest, i.e. reflect the #AC if host doesn't enable split lock
Quoted 8 lines
arch/x86/kvm/vmx/vmx.c static int handle_exception_nmi(struct kvm_vcpu *vcpu) · line 4756
Quoted 5 lines
arch/x86/kvm/vmx/vmx.c +2
-0+* detection. */ if (!split_lock_detect_on () || +guest_cpu_split_lock_detect_on (vmx) ||guest_cpu_alignment_check_enabled (vcpu)) {If the host has split lock detection disabled then how is the guest supposed to have it enabled in the first place?
So we need to reach an agreement on whether we need a state that host turns it off but feature is available to be exposed to guest.
Quoted 4 lines
Quoted 4 lines
arch/x86/kvm/vmx/vmx.c +0
-0@@ -6631,6 +6653,14 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu)*/ x86_spec_ctrl_set_guest (vmx->spec_ctrl,0 );
Quoted 8 lines
arch/x86/kvm/vmx/vmx.c static void vmx_vcpu_run(struct kvm_vcpu *vcpu) · line 6659
Quoted 4 lines
arch/x86/kvm/vmx/vmx.c +4
-0+if (static_cpu_has (X86_FEATURE_SPLIT_LOCK_DETECT ) && +guest_cpu_split_lock_detect_on (vmx)) { +if (test_thread_flag (TIF_SLD )) +sld_turn_back_on ();This is completely inconsistent behaviour. The only way that
TIF_SLDis set is when the host hassld_state==sld_warnand the guest triggered a split lock #AC.
Can you image the case that both host and guest set sld_state == sld_warn.
1. There is guest userspace thread causing split lock.
2. It sets TIF_SLD for the thread in guest, and clears SLD bit to re- execute the instruction in guest.
3. Then it still causes #AC since hardware SLD is not cleared. In host kvm, we call handle_user_split_lock() that sets TIF_SLD for this VMM thread, and clears hardware SLD bit. Then it enters guest and re-execute the instruction.
4. In guest, it schedules to another thread without TIF_SLD being set. it sets the SLD bit to detect the split lock for this thread. So for this purpose, we need to turn sld back on for the VMM thread, otherwise this guest vcpu cannot catch split lock any more.
Quoted 12 lines
'warn' means that the split lock event is registered and a printk emitted and after that the task runs with split lock detection disabled.
It does not matter at all if the task triggered the #AC while in guest or in host user space mode. Stop claiming that virt is special. The only special thing about virt is, that it is using a different mechanism to exit kernel mode. Aside of that from the kernel POV it is completely irrelevant whether the task triggered the split lock in host user space or in guest mode.
If the SLD mode is fatal, then the task is killed no matter what.
Quoted 7 lines
Please sit down and go through your patches and rethink every single line instead of sending out yet another half baken and hastily cobbled together pile.
To be clear, Patch 1 and 2 make sense on their own, so I'm tempted to pick them up right now, but the rest is going to be 5.8 material no matter what.
Alright.
Do you need me to spin a new version of patch 1 to clear SLD bit on APs if SLD_OFF?
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
Quoted 10 lines · Xiaoyao Li
Quoted 6 lines · Thomas Gleixner
arch/x86/kvm/vmx/vmx.c static int handle_exception_nmi(struct kvm_vcpu *vcpu) · line 4756
Quoted 3 lines
if (!split_lock_detect_on() ||
arch/x86/kvm/vmx/vmx.c +1
-0+ guest_cpu_split_lock_detect_on(vmx) || guest_cpu_alignment_check_enabled(vcpu)) {If the host has split lock detection disabled then how is the guest supposed to have it enabled in the first place?
So we need to reach an agreement on whether we need a state that host turns it off but feature is available to be exposed to guest.
There is a very simple agreement:
If the host turns it off, then it is not available at all
If the host sets 'warn' then this applies to everything
If the host sets 'fatal' then this applies to everything
Make it simple and consistent.
Quoted 11 lines
Quoted 8 lines
arch/x86/kvm/vmx/vmx.c static void vmx_vcpu_run(struct kvm_vcpu *vcpu) · line 6659
Quoted 4 lines
arch/x86/kvm/vmx/vmx.c +4
-0+if (static_cpu_has (X86_FEATURE_SPLIT_LOCK_DETECT ) && +guest_cpu_split_lock_detect_on (vmx)) { +if (test_thread_flag (TIF_SLD )) +sld_turn_back_on ();This is completely inconsistent behaviour. The only way that
TIF_SLDis set is when the host hassld_state==sld_warnand the guest triggered a split lock #AC.Can you image the case that both host and guest set
sld_state==sld_warn.
Quoted 11 lines
1. There is guest userspace thread causing split lock.
2. It setsTIF_SLDfor the thread in guest, and clears SLD bit to re- execute the instruction in guest.
3. Then it still causes #AC since hardware SLD is not cleared. In host kvm, we callhandle_user_split_lock()that setsTIF_SLDfor this VMM thread, and clears hardware SLD bit. Then it enters guest and re-execute the instruction.
4. In guest, it schedules to another thread withoutTIF_SLDbeing set. it sets the SLD bit to detect the split lock for this thread. So for this purpose, we need to turn sld back on for the VMM thread, otherwise this guest vcpu cannot catch split lock any more.
If you really want to address that scenario, then why are you needing any of those completely backwards interfaces at all?
Just because your KVM exception trap uses the host handling function which sets TIF_SLD?
Please sit down, go back to the drawing board and figure it out how to solve that without creating inconsistent state and duct tape functions to deal with that.
I'm not rewriting the patches for you this time.
Quoted 2 lines
Do you need me to spin a new version of patch 1 to clear SLD bit on APs if
SLD_OFF?
I assume you can answer that question yourself.
Thanks,
tglx
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
Quoted 14 lines · Thomas Gleixner
Quoted 10 lines · Xiaoyao Li
Quoted 6 lines · Thomas Gleixner
arch/x86/kvm/vmx/vmx.c static int handle_exception_nmi(struct kvm_vcpu *vcpu) · line 4756
Quoted 3 lines
if (!split_lock_detect_on() ||arch/x86/kvm/vmx/vmx.c +1
-0+ guest_cpu_split_lock_detect_on(vmx) || guest_cpu_alignment_check_enabled(vcpu)) {If the host has split lock detection disabled then how is the guest supposed to have it enabled in the first place?
So we need to reach an agreement on whether we need a state that host turns it off but feature is available to be exposed to guest.
There is a very simple agreement:
Quoted 7 lines
If the host turns it off, then it is not available at all
If the host sets 'warn' then this applies to everything
If the host sets 'fatal' then this applies to everything
Make it simple and consistent.
OK. you are the boss.
Quoted 23 lines
Quoted 11 lines
Quoted 8 lines
arch/x86/kvm/vmx/vmx.c static void vmx_vcpu_run(struct kvm_vcpu *vcpu) · line 6659
Quoted 4 lines
arch/x86/kvm/vmx/vmx.c +4
-0+if (static_cpu_has (X86_FEATURE_SPLIT_LOCK_DETECT ) && +guest_cpu_split_lock_detect_on (vmx)) { +if (test_thread_flag (TIF_SLD )) +sld_turn_back_on ();This is completely inconsistent behaviour. The only way that
TIF_SLDis set is when the host hassld_state==sld_warnand the guest triggered a split lock #AC.Can you image the case that both host and guest set
sld_state==sld_warn.Quoted 11 lines
1. There is guest userspace thread causing split lock.
2. It setsTIF_SLDfor the thread in guest, and clears SLD bit to re- execute the instruction in guest.
3. Then it still causes #AC since hardware SLD is not cleared. In host kvm, we callhandle_user_split_lock()that setsTIF_SLDfor this VMM thread, and clears hardware SLD bit. Then it enters guest and re-execute the instruction.
4. In guest, it schedules to another thread withoutTIF_SLDbeing set. it sets the SLD bit to detect the split lock for this thread. So for this purpose, we need to turn sld back on for the VMM thread, otherwise this guest vcpu cannot catch split lock any more.
Quoted 6 lines
If you really want to address that scenario, then why are you needing any of those completely backwards interfaces at all?
Just because your KVM exception trap uses the host handling function which sets
TIF_SLD?
Yes. just because KVM use the host handling function.
If you disallow me to touch codes out of kvm. It can be achieved with something like in v2:
https://lore.kernel.org/kvm/[email protected]/
Obviously re-use TIF_SLD flag to automatically switch MSR_TEST_CTRL.SLD bit when switch to/from vcpu thread is better.
And to virtualize SLD feature as full as possible for guest, we have to implement the backwards interface. If you really don't want that interface, we have to write code directly in kvm to modify TIF_SLD flag and MSR_TEST_CTRL.SLD bit.
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
Quoted 10 lines · Xiaoyao Li
Quoted 6 lines · Thomas Gleixner
If you really want to address that scenario, then why are you needing any of those completely backwards interfaces at all?
Just because your KVM exception trap uses the host handling function which sets
TIF_SLD?Yes. just because KVM use the host handling function.
If you disallow me to touch codes out of kvm. It can be achieved with
Who said you cannot touch code outside of KVM?
Quoted 2 lines
Obviously re-use
TIF_SLDflag to automatically switch MSR_TEST_CTRL.SLD bit when switch to/from vcpu thread is better.
What's better about that?
TIF_SLD has very well defined semantics. It's used to denote that the SLD bit needs to be cleared for the task when its scheduled in.
So now you overload it by clearing it magically and claim that this is better.
vCPU-thread
user space (qemu)
triggers #AC
-> exception
set TIF_SLD iotctl()
vcpu_run()
-> clear TIF_SLDIt's not better, it's simply wrong and inconsistent.
Quoted 4 lines
And to virtualize SLD feature as full as possible for guest, we have to implement the backwards interface. If you really don't want that interface, we have to write code directly in kvm to modify
TIF_SLDflag and MSR_TEST_CTRL.SLD bit.
Wrong again. KVM has absolutely no business in fiddling with TIF_SLD and the function to flip the SLD bit is simply sld_update_msr(bool on) which does not need any KVMism at all.
There are two options to handle SLD for KVM:
1) Follow strictly the host rules
If user space or guest triggered #AC then TIF_SLD is set and that task is excluded from ever setting SLD again.
2) Track KVM guest state separately
vcpu_run()
if (current_has(TIF_SLD) && guest_sld_on())
sld_update_msr(true);
else if (!current_has(TIF_SLD) && !guest_sld_on())
sld_update_msr(false);
vmenter()
....
vmexit()
if (current_has(TIF_SLD) && guest_sld_on())
sld_update_msr(false);
else if (!current_has(TIF_SLD) && !guest_sld_on())
sld_update_msr(true);
If the guest triggers #AC then this solely affects guest state
and does not fiddle with TIF_SLD.Thanks,
tglx
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
Quoted 47 lines · Thomas Gleixner
Quoted 10 lines · Xiaoyao Li
Quoted 6 lines · Thomas Gleixner
If you really want to address that scenario, then why are you needing any of those completely backwards interfaces at all?
Just because your KVM exception trap uses the host handling function which sets
TIF_SLD?Yes. just because KVM use the host handling function.
If you disallow me to touch codes out of kvm. It can be achieved with
Who said you cannot touch code outside of KVM?
Quoted 2 lines
Obviously re-use
TIF_SLDflag to automatically switch MSR_TEST_CTRL.SLD bit when switch to/from vcpu thread is better.What's better about that?
TIF_SLDhas very well defined semantics. It's used to denote that the SLD bit needs to be cleared for the task when its scheduled in.So now you overload it by clearing it magically and claim that this is better.
vCPU-thread
user space (qemu) triggers #AC -> exception set TIF_SLD iotctl() vcpu_run() -> clear TIF_SLDIt's not better, it's simply wrong and inconsistent.
Quoted 4 lines
And to virtualize SLD feature as full as possible for guest, we have to implement the backwards interface. If you really don't want that interface, we have to write code directly in kvm to modify
TIF_SLDflag and MSR_TEST_CTRL.SLD bit.Wrong again. KVM has absolutely no business in fiddling with
TIF_SLDand the function to flip the SLD bit is simplysld_update_msr(bool on)which does not need any KVMism at all.
Quoted 6 lines
There are two options to handle SLD for KVM:
1) Follow strictly the host rules
If user space or guest triggered #AC then
TIF_SLDis set and that task is excluded from ever setting SLD again.
Obviously, it's not about to virtualize SLD for guest. So we don't pick this one.
Quoted 9 lines
2) Track KVM guest state separately
vcpu_run()
if (current_has(TIF_SLD) && guest_sld_on())
sld_update_msr(true);
else if (!current_has(TIF_SLD) && !guest_sld_on())
sld_update_msr(false);
vmenter()
....Quoted 9 lines
vmexit()
if (current_has(TIF_SLD) && guest_sld_on())
sld_update_msr(false);
else if (!current_has(TIF_SLD) && !guest_sld_on())
sld_update_msr(true);
If the guest triggers #AC then this solely affects guest state
and does not fiddle with TIF_SLD.OK. So when host is sld_warn, guest's SLD value can be loaded into hardware MSR when vmenter.
I'll go with this option.
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
In reply to Thomas GleixnerQuoted 20 lines · Thomas Gleixner
Quoted 19 lines · Xiaoyao Li
arch/x86/kvm/vmx/vmx.c +8
-4+static inline bool guest_cpu_split_lock_detect_on (struct vcpu_vmx *vmx) +{ +return vmx->msr_test_ctrl &MSR_TEST_CTRL_SPLIT_LOCK_DETECT ; +} +static int handle_exception_nmi (struct kvm_vcpu *vcpu) {struct vcpu_vmx *vmx =to_vmx (vcpu);@@ -4725,12 +4746,13 @@ static int handle_exception_nmi(struct kvm_vcpu *vcpu)case AC_VECTOR :/* * Reflect #AC to the guest if it's expecting the #AC, i.e. has -* legacy alignment check enabled. Pre-check host split lock -* support to avoid the VMREADs needed to check legacy #AC, -* i.e. reflect the #AC if the only possible source is legacy -* alignment checks. +* legacy alignment check enabled or split lock detect enabled. +* Pre-check host split lock support to avoid further check of +* guest, i.e. reflect the #AC if host doesn't enable split lock
Quoted 9 lines
arch/x86/kvm/vmx/vmx.c static int handle_exception_nmi(struct kvm_vcpu *vcpu) · line 4756
Quoted 5 lines
arch/x86/kvm/vmx/vmx.c +2
-0+* detection. */ if (!split_lock_detect_on () || +guest_cpu_split_lock_detect_on (vmx) ||guest_cpu_alignment_check_enabled (vcpu)) {If the host has split lock detection disabled then how is the guest supposed to have it enabled in the first place?
It is ||
Thanks,
-Xiaoyao
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
Quoted 25 lines · Xiaoyao Li
Quoted 15 lines · Thomas Gleixner
Quoted 14 lines · Xiaoyao Li
static int handle_exception_nmi(struct kvm_vcpu *vcpu) { struct vcpu_vmx *vmx = to_vmx(vcpu);arch/x86/kvm/vmx/vmx.c +3
-4@@ -4725,12 +4746,13 @@ static int handle_exception_nmi(struct kvm_vcpu *vcpu)case AC_VECTOR :/* * Reflect #AC to the guest if it's expecting the #AC, i.e. has -* legacy alignment check enabled. Pre-check host split lock -* support to avoid the VMREADs needed to check legacy #AC, -* i.e. reflect the #AC if the only possible source is legacy -* alignment checks. +* legacy alignment check enabled or split lock detect enabled. +* Pre-check host split lock support to avoid further check of +* guest, i.e. reflect the #AC if host doesn't enable split lock Quoted 9 lines
arch/x86/kvm/vmx/vmx.c static int handle_exception_nmi(struct kvm_vcpu *vcpu) · line 4756
Quoted 5 lines
arch/x86/kvm/vmx/vmx.c +2
-0+* detection. */ if (!split_lock_detect_on () || +guest_cpu_split_lock_detect_on (vmx) ||guest_cpu_alignment_check_enabled (vcpu)) {If the host has split lock detection disabled then how is the guest supposed to have it enabled in the first place?
Quoted 1 line
It is ||
Again. If the host has it disabled, then the feature flag is OFF. So how is the hypervisor exposing it in the first place?
Thanks,
tglx
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
Quoted 26 lines · Thomas Gleixner
Quoted 25 lines · Xiaoyao Li
Quoted 24 lines · Thomas Gleixner
Quoted 9 lines · Xiaoyao Li
static int handle_exception_nmi(struct kvm_vcpu *vcpu) { struct vcpu_vmx *vmx = to_vmx(vcpu);arch/x86/kvm/vmx/vmx.c +0
-2@@ -4725,12 +4746,13 @@ static int handle_exception_nmi(struct kvm_vcpu *vcpu)case AC_VECTOR :/* * Reflect #AC to the guest if it's expecting the #AC, i.e. has -* legacy alignment check enabled. Pre-check host split lock -* support to avoid the VMREADs needed to check legacy #AC, arch/x86/kvm/vmx/vmx.c static int handle_exception_nmi(struct kvm_vcpu *vcpu) · line 4756
Quoted 10 lines
arch/x86/kvm/vmx/vmx.c +5
-2-* i.e. reflect the #AC if the only possible source is legacy -* alignment checks. +* legacy alignment check enabled or split lock detect enabled. +* Pre-check host split lock support to avoid further check of +* guest, i.e. reflect the #AC if host doesn't enable split lock +* detection. */ if (!split_lock_detect_on () || +guest_cpu_split_lock_detect_on (vmx) ||guest_cpu_alignment_check_enabled (vcpu)) {If the host has split lock detection disabled then how is the guest supposed to have it enabled in the first place?
Quoted 5 lines
Quoted 1 line
It is ||
Again. If the host has it disabled, then the feature flag is OFF. So how is the hypervisor exposing it in the first place?
So what's wrong with above code?
If the host has it disabled, !split_lock_detect_on() is true, it skips following check due to ||
I guess you want something like below?
if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK)) {
inject #AC back to guest
} else {
if (guest_alignment_check_enabled() || guest_sld_on())
inject #AC back to guest
}BTW, there is an issue in my original patch that guest_sld_on() should be checked at last.
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
Quoted 6 lines · Xiaoyao Li
On 3/26/2020 7:10 PM, Thomas Gleixner wrote:
If the host has it disabled, !split_lock_detect_on() is true, it skips following check due to ||if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK)) {
inject #AC back to guest
That'd be a regular #AC, right?
Quoted 3 lines
} else {
if (guest_alignment_check_enabled() || guest_sld_on())
inject #AC back to guestHere is clearly an else path missing.
Quoted 1 line
}
Thanks,
tglx
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
Quoted 7 lines · Thomas Gleixner
Quoted 6 lines · Xiaoyao Li
On 3/26/2020 7:10 PM, Thomas Gleixner wrote:
If the host has it disabled, !split_lock_detect_on() is true, it skips following check due to ||if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK)) {
inject #AC back to guest
and return 1;
Quoted 2 lines
That'd be a regular #AC, right?
Yes.
Quoted 3 lines
Quoted 3 lines
} else { if (guest_alignment_check_enabled() || guest_sld_on()) inject #AC back to guest
and return 1;
Quoted 1 line
Here is clearly an else path missing.
the else path is fall through.
i.e. calling handle_user_split_lock().
If cannot handle, it falls through to report #AC to user space (QEMU)
Quoted 2 lines
Quoted 1 line
}
If there is no problem with the above. So what's the problem of the original?
Re: [PATCH v6 8/8] kvm: vmx: virtualize split lock detection
Quoted 26 lines · Xiaoyao Li
Quoted 7 lines · Thomas Gleixner
Quoted 6 lines · Xiaoyao Li
On 3/26/2020 7:10 PM, Thomas Gleixner wrote:
If the host has it disabled, !split_lock_detect_on() is true, it skips following check due to ||if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK)) {
inject #AC back to guestand return 1;
Quoted 2 lines
That'd be a regular #AC, right?
Yes.
Quoted 3 lines
Quoted 3 lines
} else { if (guest_alignment_check_enabled() || guest_sld_on()) inject #AC back to guestand return 1;
Quoted 1 line
Here is clearly an else path missing.
the else path is fall through.
i.e. calling
handle_user_split_lock().
Quoted 7 lines
If cannot handle, it falls through to report #AC to user space (QEMU)
Quoted 2 lines
Quoted 1 line
}
If there is no problem with the above. So what's the problem of the original?
Probably my inability to decipher the convoluted condition.
Thanks,
tglx
Quoted 1 line · Xiaoyao Li
So sorry for the noise that I forgot to CC the maillist.
You also forgot to add RESEND :-) Tagging the patches like so
[PATCH RESEND v6 0/8] x86/split_lock: Fix and virtualization of split lock detection
lets folks that received the originals identify and respond to the "new" thread.