Re: [PATCH 0/3] riscv: set vstart_eq_zero on mark_vs_dirty

[Daniel Henrique Barboza]

Hi,

Please disregard this version. I just posted a v2 after feedback from people
in Gitlab.

Thanks,

Daniel

On 2/16/24 08:28, Daniel Henrique Barboza wrote:
> Hi,
>
> This is my shot to fix https://gitlab.com/qemu-project/qemu/-/issues/1976.
>
> First patch ensures that every vector instruction that changes the
> vector state will call mark_vs_dirty(). Second patch is a trivial
> simplification.
>
> Third patch is where the bug is solved: check if 'vstart' is zeroed and
> set vstart_eq_zero accordingly.
>
> Patches based on alistair/riscv-to-apply.next. It can also be fetched
> here:
>
> https://gitlab.com/danielhb/qemu/-/tree/vstart_bug1976_v1
>
>
> Daniel Henrique Barboza (3):
>    trans_rvv.c.inc: write CSRs must call mark_vs_dirty() too
>    trans_rvv.c.inc: remove redundant mark_vs_dirty() calls
>    target/riscv/translate.c: set vstart_eq_zero in mark_vs_dirty()
>
>   target/riscv/insn_trans/trans_rvv.c.inc | 28 +++++++------------------
>   target/riscv/translate.c                | 22 +++++++++++++++++++
>   2 files changed, 29 insertions(+), 21 deletions(-)