On 3/9/22 09:46, Li RongQing wrote:
If guest kernel is configured with nopvspin, or CONFIG_PARAVIRT_SPINLOCK is disabled, or guest find its has dedicated pCPUs from realtime hint feature, the pvspinlock will be disabled, and vCPU preemption check is disabled too. but KVM still can emulating HLT for vCPU for both cases, and check if vCPU is preempted or not, and can boost performance so move the setting of pv_ops.lock.vcpu_is_preempted to kvm_guest_init, make it not depend on pvspinlock Like unixbench, single copy, vcpu with dedicated pCPU and guest kernel with nopvspin, but emulating HLT for vCPU`: Testcase Base with patch System Benchmarks Index Values INDEX INDEX Dhrystone 2 using register variables 3278.4 3277.7 Double-Precision Whetstone 822.8 825.8 Execl Throughput 1296.5 941.1 File Copy 1024 bufsize 2000 maxblocks 2124.2 2142.7 File Copy 256 bufsize 500 maxblocks 1335.9 1353.6 File Copy 4096 bufsize 8000 maxblocks 4256.3 4760.3 Pipe Throughput 1050.1 1054.0 Pipe-based Context Switching 243.3 352.0 Process Creation 820.1 814.4 Shell Scripts (1 concurrent) 2169.0 2086.0 Shell Scripts (8 concurrent) 7710.3 7576.3 System Call Overhead 672.4 673.9 ======== ======= System Benchmarks Index Score 1467.2 1483.0 Signed-off-by: Li RongQing <lirongqing@xxxxxxxxx> --- diff v3: fix building failure when CONFIG_PARAVIRT_SPINLOCK is disable and setting preemption check only when unhalt diff v2: move setting preemption check to kvm_guest_init arch/x86/kernel/kvm.c | 74 +++++++++++++++++++++++++-------------------------- 1 file changed, 37 insertions(+), 37 deletions(-) diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c index d77481ec..959f919 100644 --- a/arch/x86/kernel/kvm.c +++ b/arch/x86/kernel/kvm.c @@ -752,6 +752,39 @@ static void kvm_crash_shutdown(struct pt_regs *regs) } #endif+#ifdef CONFIG_X86_32+__visible bool __kvm_vcpu_is_preempted(long cpu) +{ + struct kvm_steal_time *src = &per_cpu(steal_time, cpu); + + return !!(src->preempted & KVM_VCPU_PREEMPTED); +} +PV_CALLEE_SAVE_REGS_THUNK(__kvm_vcpu_is_preempted); + +#else + +#include <asm/asm-offsets.h> + +extern bool __raw_callee_save___kvm_vcpu_is_preempted(long); + +/* + * Hand-optimize version for x86-64 to avoid 8 64-bit register saving and + * restoring to/from the stack. + */ +asm( +".pushsection .text;" +".global __raw_callee_save___kvm_vcpu_is_preempted;" +".type __raw_callee_save___kvm_vcpu_is_preempted, @function;" +"__raw_callee_save___kvm_vcpu_is_preempted:" +"movq __per_cpu_offset(,%rdi,8), %rax;" +"cmpb $0, " __stringify(KVM_STEAL_TIME_preempted) "+steal_time(%rax);" +"setne %al;" +"ret;" +".size __raw_callee_save___kvm_vcpu_is_preempted, .-__raw_callee_save___kvm_vcpu_is_preempted;" +".popsection"); + +#endif + static void __init kvm_guest_init(void) { int i; @@ -764,6 +797,10 @@ static void __init kvm_guest_init(void) if (kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) { has_steal_clock = 1; static_call_update(pv_steal_clock, kvm_steal_clock); + + if (kvm_para_has_feature(KVM_FEATURE_PV_UNHALT)) + pv_ops.lock.vcpu_is_preempted = + PV_CALLEE_SAVE(__kvm_vcpu_is_preempted); }
Is it necessary to check PV_UNHALT? The bit is present anyway in the steal time struct, unless it's a very old kernel. And it's safe to always return zero if the bit is not present.
Paolo
