On Wed, Aug 02, 2023 at 12:04:25PM +0300, Andrew Jones wrote: > On Tue, Aug 01, 2023 at 07:26:20PM -0300, Daniel Henrique Barboza wrote: > > Hi, > > > > In this new version 3 new patches (6, 7, 8) were added by Andrew's > > request during the v1 review. > > > > We're now avoiding throwing an -EBUSY error if a reg write is done after > > the vcpu started spinning if the value being written is the same as KVM > > already uses. This follows the design choice made in patch 3, allowing > > for userspace 'lazy write' of registers. > > > > I decided to add 3 patches instead of one because the no-op check made > > in patches 6 and 8 aren't just a matter of doing reg_val = host_val. > > They can be squashed in a single patch if required. > > > > Please check the version 1 cover-letter [1] for the motivation behind > > this work. Patches were based on top of riscv_kvm_queue. > > > > Changes from v1: > > - patches 6,7, 8 (new): > > - make reg writes a no-op, regardless of vcpu->arch.ran_atleast_once > > state, if the value being written is the same as the host > > - v1 link: https://lore.kernel.org/kvm/20230731120420.91007-1-dbarboza@xxxxxxxxxxxxxxxx/ > > > > [1] https://lore.kernel.org/kvm/20230731120420.91007-1-dbarboza@xxxxxxxxxxxxxxxx/ > > > > I found three missing conversions, which are in the diff below. Also, I > saw that the vector registers were lacking good error returns, so I reworked > that and attached a completely (not even compile) tested patch for them. ^un
