Re: [PATCH v3 3/3] target/riscv/kvm: add missing KVM CSRs

[Andrew Jones <ajones@ventanamicro.com>]

On Thu, Mar 20, 2025 at 07:25:07AM -0700, Andrea Bolognani wrote:
> On Mon, Mar 03, 2025 at 01:46:53PM +1000, Alistair Francis wrote:
> > On Mon, Feb 24, 2025 at 10:32 PM Daniel Henrique Barboza <dbarboza@ventanamicro.com> wrote:
> > > We're missing scounteren and senvcfg CSRs, both already present in the
> > > KVM UAPI.
> > >
> > > Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
> > > Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
> >
> > Acked-by: Alistair Francis <alistair.francis@wdc.com>
> 
> This patch seems to have broken KVM acceleration for me:
> 
>   $ ./build/qemu-system-riscv64 -display none -M virt,accel=kvm -cpu host
>   qemu-system-riscv64: Failed to put registers after init: No such
> file or directory
> 
> Reverting it makes QEMU work again.
> 
> My host is a SiFive HiFive Premier P550 board running Fedora 41. Note
> that, since the upstreaming effort for this SoC has just recently
> started, I'm using the 6.6-based vendor kernel.

Ancient :-)

> 
> Perhaps the KVM UAPI additions mentioned in the commit message are
> more recent than that, and we need to make QEMU's use of them
> conditional rather than unconditional?

scounteren has been around since the dawn of riscv kvm, but senvcfg has
only been there since 6.7 (just missed your ancient cut-off).

The true fix for this is to start using get-reg-list, which should
hopefully work with the 6.6 kernel too since get-reg-list support has
been around since 6.6.

A quick fix for this is to just drop senvcfg for now since nobody
noticed it was missing before (well, I noticed it was missing, but by
inspection, not test).

Thanks,
drew