Re: [PATCH v13 06/12] KVM: x86: Add Intel PT virtualization work mode
From: Jim Mattson
Date: Wed Oct 24 2018 - 12:20:31 EST
On Wed, Oct 24, 2018 at 1:05 AM, Luwei Kang <luwei.kang@xxxxxxxxx> wrote:
> From: Chao Peng <chao.p.peng@xxxxxxxxxxxxxxx>
> Intel Processor Trace virtualization can be work in one
> of 2 possible modes:
> a. System-Wide mode (default):
> When the host configures Intel PT to collect trace packets
> of the entire system, it can leave the relevant VMX controls
> clear to allow VMX-specific packets to provide information
> across VMX transitions.
> KVM guest will not aware this feature in this mode and both
> host and KVM guest trace will output to host buffer.
> b. Host-Guest mode:
> Host can configure trace-packet generation while in
> VMX non-root operation for guests and root operation
> for native executing normally.
> Intel PT will be exposed to KVM guest in this mode, and
> the trace output to respective buffer of host and guest.
> In this mode, tht status of PT will be saved and disabled
> before VM-entry and restored after VM-exit if trace
> a virtual machine.
> Signed-off-by: Chao Peng <chao.p.peng@xxxxxxxxxxxxxxx>
> Signed-off-by: Luwei Kang <luwei.kang@xxxxxxxxx>
> +#define SECONDARY_EXEC_PT_USE_GPA 0x01000000
> +#define VM_EXIT_CLEAR_IA32_RTIT_CTL 0x02000000
> +#define VM_ENTRY_LOAD_IA32_RTIT_CTL 0x00040000
Where are all of these bits documented? I'm looking at the latest SDM,
volume 3 (325384-067US), and none of these bits aredocumented there.
> + GUEST_IA32_RTIT_CTL = 0x00002814,
> + GUEST_IA32_RTIT_CTL_HIGH = 0x00002815,
Where is this VMCS field documented?
> +/* Default is SYSTEM mode. */
> +static int __read_mostly pt_mode = PT_MODE_SYSTEM;
> +module_param(pt_mode, int, S_IRUGO);
As a module parameter, this doesn't allow much flexibility. Is it
possible to make this decision per-VM, using a VM capability that can
be set by userspace? (In that case, it may make sense to have a module
parameter which allows/disallows the per-VM capability.)
> +static inline bool cpu_has_vmx_intel_pt(void)
> + u64 vmx_msr;
> + rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
> + return !!(vmx_msr & MSR_IA32_VMX_MISC_INTEL_PT);
Instead of the rdmsr here, wouldn't it be better to cache the
IA32_VMX_MISC MSR in vmcs_config?
Nit: throughout this change, the '!!' isn't necessary when casting an
integer type to bool.