Re: [RFC v2 PATCH 2/7] KVM: Track the pid of the VM process

[Punit Agrawal]

Christoffer Dall <christoffer.dall@xxxxxxxxxx> writes:

> On Tue, Sep 06, 2016 at 04:22:17PM +0100, Punit Agrawal wrote:
>> Christoffer Dall <christoffer.dall@xxxxxxxxxx> writes:
>> 
>> > On Tue, Sep 06, 2016 at 12:07:59PM +0100, Punit Agrawal wrote:
>> >> Christoffer Dall <christoffer.dall@xxxxxxxxxx> writes:
>> >> 
>> >> > On Tue, Sep 06, 2016 at 10:51:27AM +0100, Punit Agrawal wrote:
>> >> >> Hi Christoffer,
>> >> >> 
>> >> >> Christoffer Dall <christoffer.dall@xxxxxxxxxx> writes:
>> >> >> 
>> >> >> > On Mon, Sep 05, 2016 at 05:31:32PM +0100, Punit Agrawal wrote:
>> >> >> >> Userspace tools such as perf can be used to profile individual
>> >> >> >> processes.
>> >> >> >> 
>> >> >> >> Track the PID of the virtual machine process to match profiling requests
>> >> >> >> targeted at it. This can be used to take appropriate action to enable
>> >> >> >> the requested profiling operations for the VMs of interest.
>> >> >> >> 
>> >> >> >> Signed-off-by: Punit Agrawal <punit.agrawal@xxxxxxx>
>> >> >> >> Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
>> >> >> >> Cc: "Radim KrÄmÃÅ" <rkrcmar@xxxxxxxxxx>
>> >> >> >> Cc: Christoffer Dall <christoffer.dall@xxxxxxxxxx>
>> >> >> >> Cc: Marc Zyngier <marc.zyngier@xxxxxxx>
>> >> >> >> ---
>> >> >> >>  include/linux/kvm_host.h | 1 +
>> >> >> >>  virt/kvm/kvm_main.c      | 2 ++
>> >> >> >>  2 files changed, 3 insertions(+)
>> >> >> >> 
>> >> >> >> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
>> >> >> >> index 9c28b4d..7c42c94 100644
>> >> >> >> --- a/include/linux/kvm_host.h
>> >> >> >> +++ b/include/linux/kvm_host.h
>> >> >> >> @@ -374,6 +374,7 @@ struct kvm_memslots {
>> >> >> >>  struct kvm {
>> >> >> >>  	spinlock_t mmu_lock;
>> >> >> >>  	struct mutex slots_lock;
>> >> >> >> +	struct pid *pid;
>> >> >> >>  	struct mm_struct *mm; /* userspace tied to this vm */
>> >> >> >>  	struct kvm_memslots *memslots[KVM_ADDRESS_SPACE_NUM];
>> >> >> >>  	struct srcu_struct srcu;
>> >> >> >> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
>> >> >> >> index 1950782..ab2535a 100644
>> >> >> >> --- a/virt/kvm/kvm_main.c
>> >> >> >> +++ b/virt/kvm/kvm_main.c
>> >> >> >> @@ -613,6 +613,7 @@ static struct kvm *kvm_create_vm(unsigned long type)
>> >> >> >>  	spin_lock_init(&kvm->mmu_lock);
>> >> >> >>  	atomic_inc(&current->mm->mm_count);
>> >> >> >>  	kvm->mm = current->mm;
>> >> >> >> +	kvm->pid = get_task_pid(current, PIDTYPE_PID);
>> >> >> >
>> >> >> > How dooes this deal with threading?  Is the idea that the user by
>> >> >> > specifying the main thread's pid will enable trapping for all vcpu
>> >> >> > threads belonging to that VM?
>> >> >> 
>> >> >> Yes that's correct - specifying the main thread PID will enable trapping
>> >> >> for the VM (all vcpus).
>> >> >> 
>> >> >> I am happy to move to a more suitable identifier if available.
>> >> >> 
>> >> >
>> >> > What is the 'main thread' ?
>> >> >
>> >> > Does something mandate that the VM is created by the thread group
>> >> > leader?  If not, is it not a bit strange from a user perspective, that
>> >> > you have to find the specific subthread pid that created the vm to
>> >> > enable this tracing for all vcpu threads and that the tgid doesn't work
>> >> > in this case?
>> >> 
>> >> Let me correct my terminology usage - the value recorded above (and used
>> >> to identify the VM) should be the tgid. It is confusing because 'ps'
>> >> reports it as pid.
>> >> 
>> >> I picked the value as existing KVM code already uses the PID of the
>> >> creating task (see kvm_create_vm_debugfs) to export VM statistics in
>> >> debugfs.
>> >> 
>> >> If I've got this wrong, then kvm_create_vm_debugfs also likely needs an
>> >> update.
>> >> 
>> >> What do you think?
>> >> 
>> > When you do get_task_pid(current, PIDTYPE_PID) it actually gets the
>> > kernel view of a PID which is the thead-id from userspace's point of
>> > view, right?
>> 
>> That makes sense. It seems to works here because the pid of the first
>> task is also the tgid of the group. And I reckon it's the same
>> assumption being made with debugfs code (more below).
>
> That is probably the implementation of all QEMU versions and kvmtool
> versions out there.
>
>> 
>> I've changed the first argument of the call to get_task_pid to
>> current->group_leader.
>> 
>> >
>> > I don't see why this has to be the same as the debugfs code, as there it
>> > makes potentially more sense to thread-specific, but for your case, are
>> > you not targeting the behavior that a user can do "ps aux | grep qemu"
>> > or whatever, and then set tracing for the reported PID (which is
>> > actually a tgid)?
>> 
>> The debugfs stats are not thread (vcpu) specific but for the VM.
>> 
>> Both values, debugfs and here, are being used to represent the VM to the
>> user. A mismatch in these identifiers will be very confusing.
>> 
>> If you agree, I can separately send a patch to address this for VM
>> debugfs directory as well.
>> 
>
> I don't know how the debugfs stuff is used or was intended, so I really
> can't speak for that.  It seems less weird to me with debugfs, because I
> imagine it can be used by simply looking at what exists in the debugfs
> directory and mapping that to a VM.

Ok, I'll let debugfs stuff be as is then.

>
> In your case, there's a clear expectation from the user that using the
> tgid should cover this VM, and it will be weird if that's not the
> case.

Right. Switching over to current->group_leader should avert problems if
userspace tools do things differently.

Thanks,
Punit

>
> -Christoffer
> _______________________________________________
> kvmarm mailing list
> kvmarm@xxxxxxxxxxxxxxxxxxxxx
> https://lists.cs.columbia.edu/mailman/listinfo/kvmarm