Aleksandar Markovic <aleksandar.marko...@rt-rk.com> writes:

> From: Miodrag Dinic <miodrag.di...@mips.com>
>
> While testing mttcg VP0 could get gets stuck in a loop waiting
> for other VPs to come up (which never actually happens). To fix this,
> kick VPs while they are being powered up by Cluster Power Controller
> in a async task which is triggered once the host thread is being
> spawned.
>
> Signed-off-by: Miodrag Dinic <miodrag.di...@mips.com>
> Signed-off-by: Leon Alrae <leon.al...@imgtec.com>
> Signed-off-by: Aleksandar Markovic <aleksandar.marko...@mips.com>
> ---
>  hw/misc/mips_cpc.c | 17 +++++++++++++++--
>  1 file changed, 15 insertions(+), 2 deletions(-)
>
> diff --git a/hw/misc/mips_cpc.c b/hw/misc/mips_cpc.c
> index 6d34574..105a109 100644
> --- a/hw/misc/mips_cpc.c
> +++ b/hw/misc/mips_cpc.c
> @@ -30,6 +30,14 @@ static inline uint64_t cpc_vp_run_mask(MIPSCPCState *cpc)
>      return (1ULL << cpc->num_vp) - 1;
>  }
>
> +static void mips_cpu_reset_async_work(CPUState *cs, run_on_cpu_data data)
> +{
> +    MIPSCPCState *cpc = (MIPSCPCState *) data.host_ptr;

Is MIPSCPCState part of the CPUState or a shared structure? If it is a
shared structure you may need to use safe work or have some form of
locking to prevent races.

> +
> +    cpu_reset(cs);
> +    cpc->vp_running |= 1ULL << cs->cpu_index;
> +}
> +
>  static void cpc_run_vp(MIPSCPCState *cpc, uint64_t vp_run)
>  {
>      CPUState *cs = first_cpu;
> @@ -37,8 +45,13 @@ static void cpc_run_vp(MIPSCPCState *cpc, uint64_t vp_run)
>      CPU_FOREACH(cs) {
>          uint64_t i = 1ULL << cs->cpu_index;
>          if (i & vp_run & ~cpc->vp_running) {
> -            cpu_reset(cs);
> -            cpc->vp_running |= i;
> +            /*
> +             * To avoid racing with a CPU we are just kicking off.
> +             * We do the final bit of preparation for the work in
> +             * the target CPUs context.
> +             */
> +            async_run_on_cpu(cs, mips_cpu_reset_async_work,
> +                             RUN_ON_CPU_HOST_PTR(cpc));
>          }
>      }
>  }


--
Alex Bennée

Reply via email to