Message-ID: <YQGAtxWqBhBUYWBN@google.com>
Date:   Wed, 28 Jul 2021 16:07:19 +0000
From:   David Matlack <dmatlack@...gle.com>
To:     Sean Christopherson <seanjc@...gle.com>
Cc:     Erdem Aktas <erdemaktas@...gle.com>,
        linux-kselftest@...r.kernel.org, Peter Gonda <pgonda@...gle.com>,
        Marc Orr <marcorr@...gle.com>, Sagi Shahar <sagis@...gle.com>,
        Paolo Bonzini <pbonzini@...hat.com>,
        Shuah Khan <shuah@...nel.org>,
        Andrew Jones <drjones@...hat.com>,
        Ben Gardon <bgardon@...gle.com>, Peter Xu <peterx@...hat.com>,
        Emanuele Giuseppe Esposito <eesposit@...hat.com>,
        Christian Borntraeger <borntraeger@...ibm.com>,
        Ricardo Koller <ricarkol@...gle.com>,
        Eric Auger <eric.auger@...hat.com>,
        Yanan Wang <wangyanan55@...wei.com>,
        Aaron Lewis <aaronlewis@...gle.com>,
        Jim Mattson <jmattson@...gle.com>,
        Oliver Upton <oupton@...gle.com>,
        Vitaly Kuznetsov <vkuznets@...hat.com>,
        Peter Shier <pshier@...gle.com>,
        Axel Rasmussen <axelrasmussen@...gle.com>,
        Zhenzhong Duan <zhenzhong.duan@...el.com>,
        "Maciej S. Szmigiero" <maciej.szmigiero@...cle.com>,
        Like Xu <like.xu@...ux.intel.com>,
        open list <linux-kernel@...r.kernel.org>,
        "open list:KERNEL VIRTUAL MACHINE (KVM)" <kvm@...r.kernel.org>
Subject: Re: [RFC PATCH 1/4] KVM: selftests: Add support for creating
 non-default type VMs

On Tue, Jul 27, 2021 at 08:47:44PM +0000, Sean Christopherson wrote:
> On Mon, Jul 26, 2021, David Matlack wrote:
> > On Mon, Jul 26, 2021 at 11:37:54AM -0700, Erdem Aktas wrote:
> > > Currently vm_create function only creates KVM_X86_LEGACY_VM type VMs.
> > > Changing the vm_create function to accept type parameter to create
> > > new VM types.
> > > 
> > > Signed-off-by: Erdem Aktas <erdemaktas@...gle.com>
> > > Reviewed-by: Sean Christopherson <seanjc@...gle.com>
> 
> *-by tags should not be added unless explicitly provided.  IIRC, our internal
> gerrit will convert +1 to Reviewed-by, but I don't think that's the case here.
> This applies to all patches in this series.
> 
> See "Using Reported-by:, Tested-by:, Reviewed-by:, Suggested-by: and Fixes:" in
> Documentation/process/submitting-patches.rst for more info.
> 
> > > Reviewed-by: Peter Gonda <pgonda@...gle.com>
> > > Reviewed-by: Marc Orr <marcorr@...gle.com>
> > > Reviewed-by: Sagi Shahar <sagis@...gle.com>
> > 
> > Reviewed-by: David Matlack <dmatlack@...gle.com>
> > 
> > (aside from the nit below)
> > 
> > > ---
> > >  .../testing/selftests/kvm/include/kvm_util.h  |  1 +
> > >  tools/testing/selftests/kvm/lib/kvm_util.c    | 29 +++++++++++++++++--
> > >  2 files changed, 27 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/tools/testing/selftests/kvm/include/kvm_util.h b/tools/testing/selftests/kvm/include/kvm_util.h
> > > index d53bfadd2..c63df42d6 100644
> > > --- a/tools/testing/selftests/kvm/include/kvm_util.h
> > > +++ b/tools/testing/selftests/kvm/include/kvm_util.h
> > > @@ -88,6 +88,7 @@ int vcpu_enable_cap(struct kvm_vm *vm, uint32_t vcpu_id,
> > >  void vm_enable_dirty_ring(struct kvm_vm *vm, uint32_t ring_size);
> > >  
> > >  struct kvm_vm *vm_create(enum vm_guest_mode mode, uint64_t phy_pages, int perm);
> > > +struct kvm_vm *__vm_create(enum vm_guest_mode mode, uint64_t phy_pages, int perm, int type);
> > 
> > nit: Consider using a more readable function name such as
> > vm_create_with_type().
> 
> Ha!  This is why I don't like doing internal reviews :-D

+1 :)

> 
> Erdem originally had vm_create_type(), I suggested __vm_create() as the double
> underscore scheme is more common in the kernel for cases where there's a default
> wrapper and an inner helper that implements the full API.
> 
> Convention aside, the argument againsts ...with_type() are that it doesn't scale,
> e.g. if someone adds another parameter parameter for which vm_create() provides a
> default, and it doesn't self-document the relationship between vm_create() and
> the inner helper, e.g. by convention, based on names alone I know that vm_create()
> likely is a wrapper around __vm_create().

True, although with __vm_create() is not solving the scalability
problem, it's just preventing scaling altogether (you can only have 1
wrapper function, vm_create). So if any caller wants to override one of
the defaults they have to override all of them.

I agree with you though in this case: __vm_create() is a better choice
(especially given the existence of vm_create_with_vcpus).

A better option than both (but would involve more work) would be to
create an options struct with all optional arguments. Unfortunately C
makes working with options structs a bit clumsy. But it's something to
consider as the number of options passed to __vm_create increases.

For example:

struct vm_options {
        enum vm_guest_mode mode;
        uint64_t phy_pages;
        int perm;
        int type;
};

struct kvm_vm *vm_create(const struct vm_options *options)
{
        ...
}

static const struct vm_options default_vm_options = {
  .mode = VM_MODE_DEFAULT,
  .phy_pages = DEFAULT_GUEST_PHY_PAGES,
  .perm = O_RDWR,
  .type = DEFAULT_VM_TYPE,
};

/* Create a VM with default options. */
vm = create_vm(&default_vm_options);

/* Create a VM with TDX enabled. */
struct vm_options options = default_vm_options;
options.type = VM_TYPE_TDX;
vm = create_vm(&options);

(I'm sure I ham-fisted the const stuff but you get the idea.)

I'm toying with introducing an options struct to perf_test_util as well
so this is very top of mind.

> 
> Compare that with the existing
> 
>   vm_create_default_with_vcpus()
>   vm_create_default()
>   vm_create_with_vcpus()
>   vm_create()
> 
> where the relationship between all the helpers is not immediately clear, and
> vm_create_with_vcpus() is a misnomer because it does much more than call vm_create()
> and instantiate vCPUs, e.g. it also instantiates the IRQ chip and loads the test
> into guest memory.

Powered by blists - more mailing lists