Re: [Qemu-devel] [PATCH 12/17] ppc: Fix writing to AMR/UAMOR

[Thomas Huth <thuth@redhat.com>]

On 14.03.2016 17:56, Cédric Le Goater wrote:
> From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> 
> The masks weren't chosen nor applied properly. The architecture specifies
> that writes to AMR are masked by UAMOR for PR=1, otherwise AMOR for HV=0.
> 
> The writes to UAMOR are masked by AMOR for HV=0
> 
> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> [clg: fixed gen_spr_amr() call in init_proc_book3s_64()]
> Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
> ---
>  target-ppc/translate_init.c | 78 +++++++++++++++++++++++++++++++++++----------
>  1 file changed, 61 insertions(+), 17 deletions(-)
> 
> diff --git a/target-ppc/translate_init.c b/target-ppc/translate_init.c
> index c921d9f53984..f2eb5f041ecd 100644
> --- a/target-ppc/translate_init.c
> +++ b/target-ppc/translate_init.c
> @@ -1070,30 +1070,72 @@ static void gen_spr_7xx (CPUPPCState *env)
>  
>  #ifdef TARGET_PPC64
>  #ifndef CONFIG_USER_ONLY
> -static void spr_read_uamr (DisasContext *ctx, int gprn, int sprn)
> +static void spr_write_amr(DisasContext *ctx, int sprn, int gprn)
>  {
> -    gen_load_spr(cpu_gpr[gprn], SPR_AMR);
> -    spr_load_dump_spr(SPR_AMR);
> -}
> +    TCGv t0 = tcg_temp_new();
> +    TCGv t1 = tcg_temp_new();
> +    TCGv t2 = tcg_temp_new();
>  
> -static void spr_write_uamr (DisasContext *ctx, int sprn, int gprn)
> -{
> -    gen_store_spr(SPR_AMR, cpu_gpr[gprn]);
> +    /* Note, the HV=1 PR=0 case is handled earlier by simply using
> +     * spr_write_generic for HV mode in the SPR table
> +     */
> +
> +    /* Build insertion mask into t1 based on context */
> +    if (ctx->pr) {
> +        gen_load_spr(t1, SPR_UAMOR);
> +    } else {
> +        gen_load_spr(t1, SPR_AMOR);
> +    }
> +
> +    /* Mask new bits into t2 */
> +    tcg_gen_and_tl(t2, t1, cpu_gpr[gprn]);
> +
> +    /* Load AMR and clear new bits in t0 */
> +    gen_load_spr(t0, SPR_AMR);
> +    tcg_gen_andc_tl(t0, t0, t1);
> +
> +    /* Or'in new bits and write it out */
> +    tcg_gen_or_tl(t0, t0, t2);
> +    gen_store_spr(SPR_AMR, t0);
>      spr_store_dump_spr(SPR_AMR);
> +
> +    tcg_temp_free(t0);
> +    tcg_temp_free(t1);
> +    tcg_temp_free(t2);
>  }
>  
> -static void spr_write_uamr_pr (DisasContext *ctx, int sprn, int gprn)
> +static void spr_write_uamor(DisasContext *ctx, int sprn, int gprn)
>  {
>      TCGv t0 = tcg_temp_new();
> +    TCGv t1 = tcg_temp_new();
> +    TCGv t2 = tcg_temp_new();
> +
> +    /* Note, the HV=1 case is handled earlier by simply using
> +     * spr_write_generic for HV mode in the SPR table
> +     */
>  
> +    /* Build insertion mask into t1 based on context */
> +    gen_load_spr(t1, SPR_AMOR);
> +
> +    /* Mask new bits into t2 */
> +    tcg_gen_and_tl(t2, t1, cpu_gpr[gprn]);
> +
> +    /* Load AMR and clear new bits in t0 */
>      gen_load_spr(t0, SPR_UAMOR);
> -    tcg_gen_and_tl(t0, t0, cpu_gpr[gprn]);
> -    gen_store_spr(SPR_AMR, t0);
> -    spr_store_dump_spr(SPR_AMR);
> +    tcg_gen_andc_tl(t0, t0, t1);
> +
> +    /* Or'in new bits and write it out */
> +    tcg_gen_or_tl(t0, t0, t2);
> +    gen_store_spr(SPR_UAMOR, t0);
> +    spr_store_dump_spr(SPR_UAMOR);
> +
> +    tcg_temp_free(t0);
> +    tcg_temp_free(t1);
> +    tcg_temp_free(t2);
>  }
>  #endif /* CONFIG_USER_ONLY */
>  
> -static void gen_spr_amr (CPUPPCState *env)
> +static void gen_spr_amr(CPUPPCState *env, bool has_iamr)
>  {
>  #ifndef CONFIG_USER_ONLY
>      /* Virtual Page Class Key protection */
> @@ -1101,15 +1143,17 @@ static void gen_spr_amr (CPUPPCState *env)
>       * userspace accessible, 29 is privileged.  So we only need to set
>       * the kvm ONE_REG id on one of them, we use 29 */
>      spr_register(env, SPR_UAMR, "UAMR",
> -                 &spr_read_uamr, &spr_write_uamr_pr,
> -                 &spr_read_uamr, &spr_write_uamr,
> +                 &spr_read_generic, &spr_write_amr,
> +                 &spr_read_generic, &spr_write_amr,
>                   0);
> -    spr_register_kvm(env, SPR_AMR, "AMR",
> +    spr_register_kvm_hv(env, SPR_AMR, "AMR",
>                       SPR_NOACCESS, SPR_NOACCESS,
> +                     &spr_read_generic, &spr_write_amr,
>                       &spr_read_generic, &spr_write_generic,
>                       KVM_REG_PPC_AMR, 0);
> -    spr_register_kvm(env, SPR_UAMOR, "UAMOR",
> +    spr_register_kvm_hv(env, SPR_UAMOR, "UAMOR",
>                       SPR_NOACCESS, SPR_NOACCESS,
> +                     &spr_read_generic, &spr_write_uamor,
>                       &spr_read_generic, &spr_write_generic,
>                       KVM_REG_PPC_UAMOR, 0);
>      spr_register_hv(env, SPR_AMOR, "AMOR",
> @@ -8093,7 +8137,7 @@ static void init_proc_book3s_64(CPUPPCState *env, int version)
>      case BOOK3S_CPU_POWER7:
>      case BOOK3S_CPU_POWER8:
>          gen_spr_book3s_ids(env);
> -        gen_spr_amr(env);
> +        gen_spr_amr(env, version >= BOOK3S_CPU_POWER8);
>          gen_spr_book3s_purr(env);
>          env->ci_large_pages = true;
>          break;

I think this last hunk (and thus the "has_iamr" parameter of that
function) rather belong to the next patch, since it is not used here yet.

Apart from that, the patch looks fine to me.

 Thomas