Re: [PATCH v2 1/2] selftests: KVM: align guest physical memory base address to 1GB

From: Sean Christopherson <seanjc@google.com>
To: Mingwei Zhang <mizhang@google.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>,
	kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
	David Matlack <dmatlack@google.com>,
	Jing Zhang <jingzhangos@google.com>, Peter Xu <peterx@redhat.com>,
	Ben Gardon <bgardon@google.com>
Subject: Re: [PATCH v2 1/2] selftests: KVM: align guest physical memory base address to 1GB
Date: Tue, 31 Aug 2021 16:54:56 +0000	[thread overview]
Message-ID: <YS5e4PGxu7tjiEBI@google.com> (raw)
In-Reply-To: <20210829182641.2505220-2-mizhang@google.com>

On Sun, Aug 29, 2021, Mingwei Zhang wrote:
> Existing selftest library function always allocates GPA range that aligns
> to the end of GPA address space, ie., the allocated GPA range guarantees to
> end at the last available GPA. This ends up with the fact that selftest
> programs cannot control the alignment of the base GPA. Depending on the
> size of the allocation, the base GPA may align only on a 4K based
> bounday.
> 
> The alignment of base GPA sometimes creates problems for dirty logging
> selftest where a 2MB-aligned or 1GB-aligned base GPA is needed to
> create NPT/EPT mappings for hugepages.
> 
> So, fix this issue and ensure all GPA allocation starts from a 1GB bounary
> in all architectures.
> 
> Cc: Sean Christopherson <seanjc@google.com>
> Cc: David Matlack <dmatlack@google.com>
> Cc: Jing Zhang <jingzhangos@google.com>
> Cc: Peter Xu <peterx@redhat.com>
> 
> Suggested-by: Ben Gardon <bgardon@google.com>
> Signed-off-by: Mingwei Zhang <mizhang@google.com>
> ---
>  tools/testing/selftests/kvm/lib/perf_test_util.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/tools/testing/selftests/kvm/lib/perf_test_util.c b/tools/testing/selftests/kvm/lib/perf_test_util.c
> index 0ef80dbdc116..96c30b8d6593 100644
> --- a/tools/testing/selftests/kvm/lib/perf_test_util.c
> +++ b/tools/testing/selftests/kvm/lib/perf_test_util.c
> @@ -93,10 +93,10 @@ struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int vcpus,
>  	guest_test_phys_mem = (vm_get_max_gfn(vm) - guest_num_pages) *
>  			      perf_test_args.guest_page_size;
>  	guest_test_phys_mem &= ~(perf_test_args.host_page_size - 1);
> -#ifdef __s390x__
> -	/* Align to 1M (segment size) */
> -	guest_test_phys_mem &= ~((1 << 20) - 1);
> -#endif
> +
> +	/* Align to 1G for all architectures */
> +	guest_test_phys_mem &= ~((1 << 30) - 1);

1gb may not be appropriate for all architectures and we don't want to _just_
test 1gb aligned memslots.  The alignment should be tied to the backing store,
even if the test is hardcoded to use THP, that way the alignment logic works
without modification if the backing store is changed.

I had a patch[1] that did this, let me go resurrect that series.  My series got
put on the backburner in favor of Yanan's series[2] which did a much better
job of identifying/handling the host virtual address alignment, but IIRC my
approach for handling GPA was correct.

[1] https://lore.kernel.org/kvm/20210210230625.550939-6-seanjc@google.com/
[2] https://lkml.kernel.org/r/20210330080856.14940-1-wangyanan55@huawei.com

> +
>  	pr_info("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem);
>  
>  	/* Add extra memory slots for testing */
> -- 
> 2.33.0.259.gc128427fd7-goog
> 