Re: [PATCH kvmtool v3] aarch64: Add stolen time support

[Marc Zyngier <maz@kernel.org>]

On 2022-02-22 14:28, Alexandru Elisei wrote:
> Hi,
> 
> On Tue, Feb 22, 2022 at 02:18:40PM +0000, Marc Zyngier wrote:
>> On 2022-02-22 13:58, Alexandru Elisei wrote:
>> > Hi,
>> >
>> > On Tue, Feb 22, 2022 at 10:08:30AM +0000, Sebastian Ene wrote:
>> > > This patch adds support for stolen time by sharing a memory region
>> > > with the guest which will be used by the hypervisor to store the
>> > > stolen
>> > > time information. The exact format of the structure stored by the
>> > > hypervisor is described in the ARM DEN0057A document.
>> > >
>> > > Signed-off-by: Sebastian Ene <sebastianene@google.com>
>> > > ---
>> > >  Changelog since v2:
>> > >  - Moved the AARCH64_PVTIME_* definitions from arm-common/kvm-arch.h
>> > > to
>> > >    arm64/pvtime.c as pvtime is only available for arm64.
>> > >
>> > >  Changelog since v1:
>> > >  - Removed the pvtime.h header file and moved the definitions to
>> > > kvm-cpu-arch.h
>> > >    Verified if the stolen time capability is supported before
>> > > allocating
>> > >    and mapping the memory.
>> > >
>> > >  Makefile                               |  1 +
>> > >  arm/aarch64/arm-cpu.c                  |  1 +
>> > >  arm/aarch64/include/kvm/kvm-cpu-arch.h |  1 +
>> > >  arm/aarch64/pvtime.c                   | 89
>> > > ++++++++++++++++++++++++++
>> > >  arm/kvm-cpu.c                          | 14 ++--
>> > >  5 files changed, 99 insertions(+), 7 deletions(-)
>> > >  create mode 100644 arm/aarch64/pvtime.c
>> > >
>> > > diff --git a/Makefile b/Makefile
>> > > index f251147..e9121dc 100644
>> > > --- a/Makefile
>> > > +++ b/Makefile
>> > > @@ -182,6 +182,7 @@ ifeq ($(ARCH), arm64)
>> > >  	OBJS		+= arm/aarch64/arm-cpu.o
>> > >  	OBJS		+= arm/aarch64/kvm-cpu.o
>> > >  	OBJS		+= arm/aarch64/kvm.o
>> > > +	OBJS		+= arm/aarch64/pvtime.o
>> > >  	ARCH_INCLUDE	:= $(HDRS_ARM_COMMON)
>> > >  	ARCH_INCLUDE	+= -Iarm/aarch64/include
>> > >
>> > > diff --git a/arm/aarch64/arm-cpu.c b/arm/aarch64/arm-cpu.c
>> > > index d7572b7..326fb20 100644
>> > > --- a/arm/aarch64/arm-cpu.c
>> > > +++ b/arm/aarch64/arm-cpu.c
>> > > @@ -22,6 +22,7 @@ static void generate_fdt_nodes(void *fdt, struct
>> > > kvm *kvm)
>> > >  static int arm_cpu__vcpu_init(struct kvm_cpu *vcpu)
>> > >  {
>> > >  	vcpu->generate_fdt_nodes = generate_fdt_nodes;
>> > > +	kvm_cpu__setup_pvtime(vcpu);
>> > >  	return 0;
>> > >  }
>> > >
>> > > diff --git a/arm/aarch64/include/kvm/kvm-cpu-arch.h
>> > > b/arm/aarch64/include/kvm/kvm-cpu-arch.h
>> > > index 8dfb82e..b57d6e6 100644
>> > > --- a/arm/aarch64/include/kvm/kvm-cpu-arch.h
>> > > +++ b/arm/aarch64/include/kvm/kvm-cpu-arch.h
>> > > @@ -19,5 +19,6 @@
>> > >
>> > >  void kvm_cpu__select_features(struct kvm *kvm, struct kvm_vcpu_init
>> > > *init);
>> > >  int kvm_cpu__configure_features(struct kvm_cpu *vcpu);
>> > > +void kvm_cpu__setup_pvtime(struct kvm_cpu *vcpu);
>> > >
>> > >  #endif /* KVM__KVM_CPU_ARCH_H */
>> > > diff --git a/arm/aarch64/pvtime.c b/arm/aarch64/pvtime.c
>> > > new file mode 100644
>> > > index 0000000..247e4f3
>> > > --- /dev/null
>> > > +++ b/arm/aarch64/pvtime.c
>> > > @@ -0,0 +1,89 @@
>> > > +#include "kvm/kvm.h"
>> > > +#include "kvm/kvm-cpu.h"
>> > > +#include "kvm/util.h"
>> > > +
>> > > +#include <linux/byteorder.h>
>> > > +#include <linux/types.h>
>> > > +
>> > > +#define AARCH64_PVTIME_IPA_MAX_SIZE	SZ_64K
>> > > +#define AARCH64_PVTIME_IPA_START	(ARM_MEMORY_AREA - \
>> > > +					 AARCH64_PVTIME_IPA_MAX_SIZE)
>> >
>> > This doesn't change the fact that it overlaps with KVM_PCI_MMIO_AREA,
>> > which is
>> > exposed to the guest in the DTB (see my reply to v2).
>> 
>> Yup, this is a bit of a problem, and overlapping regions are
>> a big no-no. Why can't the pvtime region be dynamically placed
>> after the RAM (after checking that there is enough space to
>> register it in the IPA space)?
> 
> In theory, is there something to stop someone from creating a VM with 
> enough
> memory to reach the end of the IPA space?

No, but we can either steal 64kB from that upper limit if that's the
case, or let the user know that stolen time is disabled because they
have been greedy...

         M.
-- 
Jazz is not dead. It just smells funny...