Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:3 as permitted sender) client-ip=2620:137:e000::3:3;
Date:   Wed, 27 Sep 2023 13:54:46 -0700
In-Reply-To: <20230807065137.3408970-4-zhaotianrui@loongson.cn>
Mime-Version: 1.0
References: <20230807065137.3408970-1-zhaotianrui@loongson.cn> <20230807065137.3408970-4-zhaotianrui@loongson.cn>
Message-ID: <ZRSWlqS3zQBSLFVK@google.com>
Subject: Re: [PATCH v2 3/4] KVM: selftests: Add ucall test support for LoongArch
From:   Sean Christopherson <seanjc@google.com>
To:     Tianrui Zhao <zhaotianrui@loongson.cn>
Cc:     Shuah Khan <shuah@kernel.org>, Paolo Bonzini <pbonzini@redhat.com>,
        linux-kernel@vger.kernel.org, kvm@vger.kernel.org,
        Vishal Annapurve <vannapurve@google.com>,
        Huacai Chen <chenhuacai@kernel.org>,
        WANG Xuerui <kernel@xen0n.name>, loongarch@lists.linux.dev,
        Peter Xu <peterx@redhat.com>,
        Vipin Sharma <vipinsh@google.com>, maobibo@loongson.cn
Content-Type: text/plain; charset="us-ascii"
Precedence: bulk

On Mon, Aug 07, 2023, Tianrui Zhao wrote:
> Add ucall test support for LoongArch. A ucall is a "hypercall to
> userspace".

Nit, can you explain why LoongArch uses MMIO to trigger ucall, and what alternatives
were considred (if any)?  The main reason for the ask is because we've tossed
around the idea of converting all architectures (except s390) to MMIO-based ucall
in order to reduce the number of "flavors" of ucall we have to worry about it.
If MMIO is the only reasonable choice for LoongArch, that's another reason to
double down on MMIO as the primary choice for ucall.

> Based-on: <20230803022138.2736430-1-zhaotianrui@loongson.cn>
> Signed-off-by: Tianrui Zhao <zhaotianrui@loongson.cn>
> ---
>  .../selftests/kvm/lib/loongarch/ucall.c       | 43 +++++++++++++++++++
>  1 file changed, 43 insertions(+)
>  create mode 100644 tools/testing/selftests/kvm/lib/loongarch/ucall.c
> 
> diff --git a/tools/testing/selftests/kvm/lib/loongarch/ucall.c b/tools/testing/selftests/kvm/lib/loongarch/ucall.c
> new file mode 100644
> index 000000000000..72868ddec313
> --- /dev/null
> +++ b/tools/testing/selftests/kvm/lib/loongarch/ucall.c
> @@ -0,0 +1,43 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * ucall support. A ucall is a "hypercall to userspace".
> + *
> + */
> +#include "kvm_util.h"
> +
> +/*
> + * ucall_exit_mmio_addr holds per-VM values (global data is duplicated by each
> + * VM), it must not be accessed from host code.
> + */
> +static vm_vaddr_t *ucall_exit_mmio_addr;
> +
> +void ucall_arch_init(struct kvm_vm *vm, vm_paddr_t mmio_gpa)
> +{
> +	vm_vaddr_t mmio_gva = vm_vaddr_unused_gap(vm, vm->page_size, KVM_UTIL_MIN_VADDR);
> +
> +	virt_map(vm, mmio_gva, mmio_gpa, 1);
> +
> +	vm->ucall_mmio_addr = mmio_gpa;
> +
> +	write_guest_global(vm, ucall_exit_mmio_addr, (vm_vaddr_t *)mmio_gva);
> +}
> +
> +void ucall_arch_do_ucall(vm_vaddr_t uc)
> +{
> +	WRITE_ONCE(*ucall_exit_mmio_addr, uc);

Another uber nit, you might want to put this in the header as a static inline to
avoid function calls.  I doubt it'll actually matter, but we've had enough weird,
hard-to-debug issues with ucall that minimizing the amount of generated code might
save some future pain.