[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20250508142842.1496099-2-rkrcmar@ventanamicro.com>
Date: Thu, 8 May 2025 16:28:41 +0200
From: Radim Krčmář <rkrcmar@...tanamicro.com>
To: kvm-riscv@...ts.infradead.org
Cc: kvm@...r.kernel.org,
linux-riscv@...ts.infradead.org,
linux-kernel@...r.kernel.org,
Anup Patel <anup@...infault.org>,
Atish Patra <atishp@...shpatra.org>,
Paul Walmsley <paul.walmsley@...ive.com>,
Palmer Dabbelt <palmer@...belt.com>,
Albert Ou <aou@...s.berkeley.edu>,
Alexandre Ghiti <alex@...ti.fr>,
Andrew Jones <ajones@...tanamicro.com>
Subject: [PATCH v2 0/2] VCPU reset fixes
Hello,
v2 abandons the idea of fixing current userspaces and adds a new
capability that uses MP_STATE to immediately trigger a KVM VCPU reset --
a single-purpose interface free of any unintended side effects is
desirable and MP_STATE is somewhat bendable, which avoids the need for a
new IOCTL. It's very dirty, though, so I'm not happy with it.
I have prepared a preview QEMU implementation [1] and only did light
testing with kvm-unit-tests as I still expect a high-level discussion.
Because it didn't take much effort on top, v2 expands the goals --
userspace can now choose the same post-reset state for all VCPUs.
(This is achieved by returning from userspace on sbi_hart_stop; let me
know if I should rather do that with another KVM capability.)
This series does not fully address the VCPU hot-unplug.
We still need a better KVM interface there, because the VCPU stopped by
sbi_hart_stop is still assumed to be present by sbi_hart_start.
I see roughly two options, with different drawbacks:
* add a new MP_STATE for the SBI HSM acceleration in KVM
* return to userspace even on sbi_hart_start and sbi_hart_get_status
(We might want to have a better interface for handing SBI in userspace
anyway, so that could be handled later.)
v2 is based on kvm-riscv/riscv_kvm_queue, 376e3c0f8aa5 ("KVM: RISC-V:
remove unnecessary SBI reset state").
Most of the patches from v1 got merged, thanks Anup, so v2 only contains
the completely redesigned approach [2/2] and a minor code style
improvement that fell through the cracks when the patches got merged out
of order [1/2].
---
1: https://github.com/qemu/qemu/commit/dec042841f383c6a825b1642b86d9f585778a2e7
Radim Krčmář (2):
KVM: RISC-V: reset smstateen in a better place
RISC-V: KVM: add KVM_CAP_RISCV_MP_STATE_RESET
Documentation/virt/kvm/api.rst | 15 +++++++++++
arch/riscv/include/asm/kvm_host.h | 3 +++
arch/riscv/include/asm/kvm_vcpu_sbi.h | 1 +
arch/riscv/kvm/vcpu.c | 39 ++++++++++++++++-----------
arch/riscv/kvm/vcpu_sbi.c | 17 ++++++++++++
arch/riscv/kvm/vcpu_sbi_hsm.c | 7 ++++-
arch/riscv/kvm/vcpu_sbi_system.c | 3 ++-
arch/riscv/kvm/vm.c | 13 +++++++++
include/uapi/linux/kvm.h | 1 +
9 files changed, 82 insertions(+), 17 deletions(-)
--
2.49.0
Powered by blists - more mailing lists